cuCIM API Reference#
Clara Submodules#
- class cucim.clara.CuImage#
- Attributes
associated_imagesReturns a set of associated image names.
channel_namesA channel name list.
coord_sysCoordinate frame in which the direction cosines are measured.
deviceA device type.
dimsA string containing a list of dimensions being requested.
directionDirection cosines (size is always 3x3).
dtypeThe data type of the image.
is_loadedTrue if image data is loaded & available.
metadataA metadata object as dict.
ndimThe number of dimensions.
originPhysical location of (0, 0, 0) (size is always 3).
pathUnderlying file path for this object.
raw_metadataA raw metadata string.
resolutionsReturns a dict that includes resolution information.
shapeA tuple of dimension sizes (in the order of dims)
typestrThe data type of the image in string format.
Methods
associated_image(self[, name, device])Returns an associated image for the given name, as a CuImage object.
cache([type])Get cache object.
close(self)Closes the file handle.
profiler(**kwargs)Get profiler object.
read_region(self[, location, size, level, ...])Returns a subresolution image.
save(self, arg0)Saves image data to the file path.
size(self[, dim_order])Returns size as a tuple for the given dimension order.
spacing(self[, dim_order])Returns physical size in tuple.
spacing_units(self[, dim_order])Units for each spacing element (size is same with ndim).
- associated_image(self: cucim.clara._cucim.CuImage, name: str = '', device: cucim.clara._cucim.io.Device = cpu) object#
Returns an associated image for the given name, as a CuImage object.
- property associated_images#
Returns a set of associated image names.
Digital Pathology image usually has a label/thumbnail or a macro image(low-power snapshot of the entire glass slide). Names of those images (such as ‘macro’ and ‘label’) are in associated_images.
- static cache(type: object = None, **kwargs) cucim.clara._cucim.cache.ImageCache#
Get cache object.
- property channel_names#
A channel name list.
- close(self: cucim.clara._cucim.CuImage) None#
Closes the file handle.
Once the file handle is closed, the image object (if loaded before) still exists but cannot read additional images from the file.
- property coord_sys#
Coordinate frame in which the direction cosines are measured.
Available Coordinate frame names are not finalized yet.
- property device#
A device type.
By default t is cpu (It will be changed since v0.19.0).
- property dims#
A string containing a list of dimensions being requested.
- The default is to return the six standard dims (‘STCZYX’) unless it is a DP multi-resolution image.
[sites, time, channel(or wavelength), z, y, x]. S - Sites or multiposition locations.
NOTE: in OME-TIFF’s metadata, dimension order would be specified as ‘XYZCTS’ (first one is fast-iterating dimension).
- property direction#
Direction cosines (size is always 3x3).
- property dtype#
The data type of the image.
- property is_loaded#
True if image data is loaded & available.
- is_trace_enabled = False#
- property metadata#
A metadata object as dict.
It would be a dictionary(key-value pair) in general but can be a complex object (e.g., OME-TIFF metadata).
- property ndim#
The number of dimensions.
- property origin#
Physical location of (0, 0, 0) (size is always 3).
- property path#
Underlying file path for this object.
- static profiler(**kwargs) cucim.clara._cucim.profiler.Profiler#
Get profiler object.
- property raw_metadata#
A raw metadata string.
- read_region(self: cucim.clara._cucim.CuImage, location: Iterable = (), size: List[int] = (), level: int = 0, num_workers: int = 0, batch_size: int = 1, drop_last: bool = False, prefetch_factor: int = 2, shuffle: bool = False, seed: int = 0, device: cucim.clara._cucim.io.Device = cpu, buf: object = None, shm_name: str = '', **kwargs) object#
Returns a subresolution image.
location and size’s dimension order is reverse of image’s dimension order.
Need to specify (X,Y) and (Width, Height) instead of (Y,X) and (Height, Width).
If location is not specified, location would be (0, 0) if Z=0. Otherwise, location would be (0, 0, 0)
Like OpenSlide, location is level-0 based coordinates (using the level-0 reference frame)
If size is not specified, size would be (width, height) of the image at the specified level.
<not supported yet> Additional parameters (S,T,C,Z) are similar to <https://allencellmodeling.github.io/aicsimageio/aicsimageio.html#aicsimageio.aics_image.AICSImage.get_image_data>
Do not yet support indices/ranges for (S,T,C,Z).
Default value for level, S, T, Z are zero.
Default value for C is -1 (whole channels)
<not supported yet> device could be one of the following strings or Device object: e.g., ‘cpu’, ‘cuda’, ‘cuda:0’ (use index 0), cucim.clara.io.Device(cucim.clara.io.CUDA,0).
<not supported yet> If buf is specified (buf’s type can be either numpy object that implements __array_interface__, or cupy-compatible object that implements __cuda_array_interface__), the read image would be saved into buf object without creating CPU/GPU memory.
<not supported yet> If shm_name is specified, shared memory would be created and data would be read in the shared memory.
- property resolutions#
Returns a dict that includes resolution information.
level_count: The number of levels
level_dimensions: A tuple of dimension tuples (width, height)
level_downsamples: A tuple of down-sample factors
level_tile_sizes: A tuple of tile size tuple (tile width, tile_height)
- save(self: cucim.clara._cucim.CuImage, arg0: str) None#
Saves image data to the file path.
Currently it supports only .ppm file format that can be viewed by eog command in Ubuntu.
- property shape#
A tuple of dimension sizes (in the order of dims)
- size(self: cucim.clara._cucim.CuImage, dim_order: str = '') List[int]#
Returns size as a tuple for the given dimension order.
- spacing(self: cucim.clara._cucim.CuImage, dim_order: str = '') List[float]#
Returns physical size in tuple.
If dim_order is specified, it returns phisical size for the dimensions. If a dimension given by the dim_order doesn’t exist, it returns 1.0 by default for the missing dimension.
- Args:
dim_order: A dimension string (e.g., ‘XYZ’)
- Returns:
A tuple with physical size for each dimension
- spacing_units(self: cucim.clara._cucim.CuImage, dim_order: str = '') List[str]#
Units for each spacing element (size is same with ndim).
- property typestr#
The data type of the image in string format.
The value can be converted to NumPy’s dtype using numpy.dtype(). (e.g., numpy.dtype(img.typestr)).
- class cucim.clara.DLDataType#
- Attributes
- property bits#
Number of bits, common choices are 8, 16, 32.
- property code#
Type code of base types.
- property lanes#
Number of lanes in the type, used for vector types.
- class cucim.clara.DLDataTypeCode#
Members:
DLInt
DLUInt
DLFloat
DLBfloat
- Attributes
namename(self: handle) -> str
- value
- DLBfloat = <DLDataTypeCode.DLBfloat: 4>#
- DLFloat = <DLDataTypeCode.DLFloat: 2>#
- DLInt = <DLDataTypeCode.DLInt: 0>#
- DLUInt = <DLDataTypeCode.DLUInt: 1>#
- property name#
- property value#
cache#
- class cucim.clara.cache.CacheType#
Members:
NoCache
PerProcess
SharedMemory
- Attributes
namename(self: handle) -> str
- value
- NoCache = <CacheType.NoCache: 0>#
- PerProcess = <CacheType.PerProcess: 1>#
- property name#
- property value#
- class cucim.clara.cache.ImageCache#
- Attributes
capacityA capacity of list/hashmap.
configReturns the dictionary of configuration.
free_memoryA cache memory size available in the cache memory.
hit_countA cache hit count.
memory_capacityA capacity of cache memory.
memory_sizeA size of cache memory used.
miss_countA cache miss count.
sizeA size of list/hashmap.
typeA Cache type.
Methods
record(self[, value])Records the cache statistics.
reserve(self, memory_capacity, **kwargs)Reserves more memory if possible.
- property capacity#
A capacity of list/hashmap.
- property config#
Returns the dictionary of configuration.
- property free_memory#
A cache memory size available in the cache memory.
- property hit_count#
A cache hit count.
- property memory_capacity#
A capacity of cache memory.
- property memory_size#
A size of cache memory used.
- property miss_count#
A cache miss count.
- record(self: cucim.clara._cucim.cache.ImageCache, value: object = None) bool#
Records the cache statistics.
- reserve(self: cucim.clara._cucim.cache.ImageCache, memory_capacity: int, **kwargs) None#
Reserves more memory if possible.
- property size#
A size of list/hashmap.
- property type#
A Cache type.
- cucim.clara.cache.preferred_memory_capacity(img: object = None, image_size: Optional[List[int]] = None, tile_size: Optional[List[int]] = None, patch_size: Optional[List[int]] = None, bytes_per_pixel: int = 3) int#
Returns a good cache memory capacity value in MiB for the given conditions.
Please see how the value is calculated: https://godbolt.org/z/8vxnPfKM5
- Args:
img: A CuImage object that can provide image_size, tile_size, bytes_per_pixel information. If this argument is provided, only patch_size from the arguments is used for the calculation. image_size: A list of values that presents the image size (width, height). tile_size: A list of values that presents the tile size (width, height). The default value is (256, 256). patch_size: A list of values that presents the patch size (width, height). The default value is (256, 256). bytes_per_pixel: The number of bytes that each pixel in the 2D image takes place. The default value is 3.
- Returns:
int: The suggested memory capacity in MiB.
filesystem#
- class cucim.clara.filesystem.CuFileDriver#
Methods
close(self)Closes opened file if not closed.
pread(self, buf, count, file_offset[, ...])Reads up to count bytes from the file driver at offset file_offset (from the start of the file) into the buffer buf starting at offset buf_offset.
pwrite(self, buf, count, file_offset[, ...])Reads up to count bytes from the file driver at offset file_offset (from the start of the file) into the buffer buf starting at offset buf_offset.
- close(self: cucim.clara._cucim.filesystem.CuFileDriver) bool#
Closes opened file if not closed.
- pread(self: cucim.clara._cucim.filesystem.CuFileDriver, buf: object, count: int, file_offset: int, buf_offset: int = 0) int#
Reads up to count bytes from the file driver at offset file_offset (from the start of the file) into the buffer buf starting at offset buf_offset. The file offset is not changed.
- Args:
buf: A buffer where read bytes are stored. Buffer can be either in CPU memory or (CUDA) GPU memory. count: The number of bytes to read. file_offset: An offset from the start of the file. buf_offset: An offset from the start of the buffer. Default value is 0.
- Returns:
The number of bytes read if succeed, -1 otherwise.
- pwrite(self: cucim.clara._cucim.filesystem.CuFileDriver, buf: object, count: int, file_offset: int, buf_offset: int = 0) int#
Reads up to count bytes from the file driver at offset file_offset (from the start of the file) into the buffer buf starting at offset buf_offset. The file offset is not changed.
- Args:
buf: A buffer where write bytes come from. Buffer can be either in CPU memory or (CUDA) GPU memory. count: The number of bytes to write. file_offset: An offset from the start of the file. buf_offset: An offset from the start of the buffer. Default value is 0.
- Returns:
The number of bytes written if succeed, -1 otherwise.
- cucim.clara.filesystem.close(arg0: cucim.clara._cucim.filesystem.CuFileDriver) bool#
Closes the given file driver.
- Args:
fd: An CuFileDriver object.
- Returns:
True if succeed, False otherwise.
- cucim.clara.filesystem.discard_page_cache(file_path: str) bool#
Discards a system (page) cache for the given file path.
- Args:
file_path: A file path to drop system cache.
- Returns:
True if succeed, False otherwise.
- cucim.clara.filesystem.open(file_path: str, flags: str, mode: int = 420) cucim.clara._cucim.filesystem.CuFileDriver#
Open file with specific flags and mode.
‘flags’ can be one of the following flag string:
“r”: os.O_RDONLY
“r+”: os.O_RDWR
“w”: os.O_RDWR | os.O_CREAT | os.O_TRUNC
“a”: os.O_RDWR | os.O_CREAT
In addition to above flags, the method append os.O_CLOEXEC and os.O_DIRECT by default.
The following is optional flags that can be added to above string:
‘p’: Use POSIX APIs only (first try to open with O_DIRECT). It does not use GDS.
‘n’: Do not add O_DIRECT flag.
‘m’: Use memory-mapped file. This flag is supported only for the read-only file descriptor.
When ‘m’ is used, PROT_READ and MAP_SHARED are used for the parameter of mmap() function.
- Args:
file_path: A file path to open. flags: File flags in string. Default value is “r”. mode: A file mode. Default value is ‘0o644’.
- Returns:
An object of CuFileDriver.
- cucim.clara.filesystem.pread(fd: cucim.clara._cucim.filesystem.CuFileDriver, buf: object, count: int, file_offset: int, buf_offset: int = 0) int#
Reads up to count bytes from file driver fd at offset offset (from the start of the file) into the buffer buf starting at offset buf_offset. The file offset is not changed.
- Args:
fd: An object of CuFileDriver. buf: A buffer where read bytes are stored. Buffer can be either in CPU memory or (CUDA) GPU memory. count: The number of bytes to read. file_offset: An offset from the start of the file. buf_offset: An offset from the start of the buffer. Default value is 0.
- Returns:
The number of bytes read if succeed, -1 otherwise.
- cucim.clara.filesystem.pwrite(fd: cucim.clara._cucim.filesystem.CuFileDriver, buf: object, count: int, file_offset: int, buf_offset: int = 0) int#
Write up to count bytes from the buffer buf starting at offset buf_offset to the file driver fd at offset offset (from the start of the file). The file offset is not changed.
- Args:
fd: An object of CuFileDriver. buf: A buffer where write bytes come from. Buffer can be either in CPU memory or (CUDA) GPU memory. count: The number of bytes to write. file_offset: An offset from the start of the file. buf_offset: An offset from the start of the buffer. Default value is 0.
- Returns:
The number of bytes written if succeed, -1 otherwise.
io#
- class cucim.clara.io.Device#
-
Methods
parse_type(arg0)Create DeviceType object from the device name string.
- property index#
Device index.
- static parse_type(arg0: str) cucim.clara._cucim.io.DeviceType#
Create DeviceType object from the device name string.
- property type#
Device type.
- class cucim.clara.io.DeviceType#
Members:
CPU
CUDA
CUDAHost
CUDAManaged
CPUShared
CUDAShared
- Attributes
namename(self: handle) -> str
- value
- CPU = <DeviceType.CPU: 1>#
- CUDA = <DeviceType.CUDA: 2>#
- CUDAHost = <DeviceType.CUDAHost: 3>#
- CUDAManaged = <DeviceType.CUDAManaged: 13>#
- property name#
- property value#
core Submodules#
color#
- cucim.core.operations.color.color_jitter(img: Any, brightness=0, contrast=0, saturation=0, hue=0)#
Applies color jitter by random sequential application of 4 operations (brightness, contrast, saturation, hue).
- Parameters
- imgchannel first, cupy.ndarray or numpy.ndarray
Input data of shape (C, H, W). Can also batch process input of shape (N, C, H, W). Can be a numpy.ndarray or cupy.ndarray.
- brightnessfloat or 2-tuple of float, optional
Non-negative factor to jitter the brightness by. When brightness is a scalar, scaling will be by a random value in range
[max(0, 1 - brightness), (1 + brightness)]. brightness can also be a 2-tuple specifying the range for the random scaling factor. A value of 0 or (1, 1) will result in no change.- contrastfloat or 2-tuple of float, optional
Non-negative factor to jitter the contrast by. When contrast is a scalar, scaling will be by a random value between
[max(0, 1 - contrast), (1 + contrast)]. contrast can also be a 2-tuple specifying the range for the random scaling factor. A value of 0 or (1, 1) will result in no change.- saturationfloat or 2-tuple of float, optional
Non-negative factor to jitter the saturation by. When saturation is a scalar, scaling will be by a random value between
[max(0, 1 - saturation), (1 + saturation)]. saturation can also be a 2-tuple specifying the range for the random scaling factor. A value of 0 or (1, 1) will result in no change.- huefloat or 2-tuple of float, optional
Factor between [-0.5, 0.5] to jitter hue by. When hue is a scalar, scaling will be by a random value between in the range
[-hue, hue]. hue can also be a 2-tuple specifying the range. A value of 0 or (0, 0) will result in no change.
- Returns
- outcupy.ndarray or numpy.ndarray
Output data. Same dimensions and type as input.
- Raises
- ValueError
If ‘brightness’,’contrast’,’saturation’ or ‘hue’ is outside of allowed range
- TypeError
If input ‘img’ is not cupy.ndarray or numpy.ndarray
Examples
>>> import cucim.core.operations.color as ccl >>> # input is channel first 3d array >>> output_array = ccl.color_jitter(input_arr,.25,.75,.25,.04)
- cucim.core.operations.color.image_to_absorbance(image, source_intensity=255.0, dtype=<class 'numpy.float32'>)#
Convert an image to units of absorbance (optical density).
- Parameters
- imagendarray
The image to convert to absorbance. Can be single or multichannel.
- source_intensityfloat, optional
Reference intensity for image.
- dtypenumpy.dtype, optional
The floating point precision at which to compute the absorbance.
- Returns
- absorbancendarray
The absorbance computed from image.
Notes
If image has an integer dtype it will be clipped to range
[1, source_intensity], while float image inputs are clipped to range[source_intensity/255, source_intensity]. The minimum is to avoid log(0). Absorbance is then given byabsorbance = log(image / source_intensity).
- cucim.core.operations.color.normalize_colors_pca(image, source_intensity: float = 240.0, alpha: float = 1.0, beta: float = 0.345, ref_stain_coeff: Union[tuple, ndarray] = ((0.5626, 0.2159), (0.7201, 0.8012), (0.4062, 0.5581)), ref_max_conc: Union[tuple, ndarray] = (1.9705, 1.0308), image_type: str = 'intensity', channel_axis: int = 0)#
Extract the matrix of stain coefficient from the image.
- Parameters
- imagenp.ndarray
RGB image to determine concentrations for. Intensities should typically be within unsigned 8-bit integer intensity range ([0, 255]) when
image_type == "intensity".- source_intensityfloat, optional
Transmitted light intensity. The algorithm will clip image intensities above the specified source_intensity and then normalize by source_intensity so that image intensities are <= 1.0. Only used when image_type==”intensity”.
- alphafloat, optional
Algorithm parameter controlling the
[alpha, 100 - alpha]percentile range used as a robust [min, max] estimate.- betafloat, optional
Absorbance (optical density) threshold below which to consider pixels as transparent. Transparent pixels are excluded from the estimation.
- ref_stain_coeffarray-like
Reference stain coefficients as determined by the output of stain_extraction_pca for a reference image.
- ref_max_conctuple or cp.ndarray
The reference maximum concentrations.
- image_type{“intensity”, “absorbance”}, optional
With the default image_type of “intensity”, the image will be transformed to an absorbance using
image_to_absorbance. If the input image is already an absorbance image, then image_type should be set to “absorbance” instead.- channel_axisint, optional
The axis corresponding to color channels (default is the last axis).
- Returns
- stain_coeffnp.ndarray
Stain attenuation coefficient matrix derived from the image, where the first column corresponds to H, the second column is E and the rows are RGB values.
Notes
The default beta of 0.345 is equivalent to the use of 0.15 in [1]. The difference is due to our use of the natural log instead of a decadic log (log10) when computing the absorbance.
References
- 1
M. Macenko et al., “A method for normalizing histology slides for quantitative analysis,” 2009 IEEE International Symposium on Biomedical Imaging: From Nano to Macro, 2009, pp. 1107-1110, doi: 10.1109/ISBI.2009.5193250. http://wwwx.cs.unc.edu/~mn/sites/default/files/macenko2009.pdf
- cucim.core.operations.color.stain_extraction_pca(image, source_intensity=240, alpha=1, beta=0.345, *, channel_axis=0, image_type='intensity')#
Extract the matrix of H & E stain coefficient from an image.
Uses a method that selects stain vectors based on the angle distribution within a best-fit plane determined by principle component analysis (PCA) [1].
- Parameters
- imagecp.ndarray
RGB image to perform stain extraction on. Intensities should typically be within unsigned 8-bit integer intensity range ([0, 255]) when
image_type == "intensity".- source_intensityfloat, optional
Transmitted light intensity. The algorithm will clip image intensities above the specified source_intensity and then normalize by source_intensity so that image intensities are <= 1.0. Only used when image_type==”intensity”.
- alphafloat, optional
Algorithm parameter controlling the
[alpha, 100 - alpha]percentile range used as a robust [min, max] estimate.- betafloat, optional
Absorbance (optical density) threshold below which to consider pixels as transparent. Transparent pixels are excluded from the estimation.
- Returns
- stain_coeffcp.ndarray
Stain attenuation coefficient matrix derived from the image, where the first column corresponds to H, the second column is E and the rows are RGB values.
Notes
The default beta of 0.345 is equivalent to the use of 0.15 in [1]. The difference is due to our use of the natural log instead of a decadic log (log10) when computing the absorbance.
References
- 1(1,2)
M. Macenko et al., “A method for normalizing histology slides for quantitative analysis,” 2009 IEEE International Symposium on Biomedical Imaging: From Nano to Macro, 2009, pp. 1107-1110, doi: 10.1109/ISBI.2009.5193250. http://wwwx.cs.unc.edu/~mn/sites/default/files/macenko2009.pdf
expose#
intensity#
- cucim.core.operations.intensity.normalize_data(img: Any, norm_constant: float, min_value: float, max_value: float, type: str = 'range') Any#
Apply intensity normalization to the input array. Normalize intensities to the range of [0, norm_constant].
- Parameters
- imgchannel first, cupy.ndarray or numpy.ndarray
Input data of shape (C, H, W). Can also batch process input of shape (N, C, H, W). Can be a numpy.ndarray or cupy.ndarray.
- norm_constant: float
Normalization range of the input data. [0, norm_constant]
- min_valuefloat
Minimum intensity value in input data.
- max_valuefloat
Maximum intensity value in input data.
- type{‘range’, ‘atan’}
Type of normalization.
- Returns
- outcupy.ndarray or numpy.ndarray
Output data. Same dimensions and type as input.
- Raises
- TypeError
If input ‘img’ is not cupy.ndarray or numpy.ndarray
- ValueError
If input original intensity min and max are same
- ValueError
If incorrect normalization type is invoked
Examples
>>> import cucim.core.operations.intensity as its >>> # input is channel first 3d array >>> output_array = its.normalize_data(input_arr, 10, 0 , 255)
- cucim.core.operations.intensity.rand_zoom(img: Any, min_zoom: Union[Sequence[float], float] = 0.9, max_zoom: Union[Sequence[float], float] = 1.1, prob: float = 0.1, whole_batch: bool = False)#
Randomly Calls zoom with random zoom factor
- Parameters
- imgchannel first, cupy.ndarray or numpy.ndarray
Input data of shape (C, H, W). Can also batch process input of shape (N, C, H, W). Can be a numpy.ndarray or cupy.ndarray.
- min_zoom: Min zoom factor. Can be float or sequence same size as image.
If a float, select a random factor from [min_zoom, max_zoom] then apply to all spatial dims to keep the original spatial shape ratio. If a sequence, min_zoom should contain one value for each spatial axis. If 2 values provided for 3D data, use the first value for both H & W dims to keep the same zoom ratio.
- max_zoom: Max zoom factor. Can be float or sequence same size as image.
If a float, select a random factor from [min_zoom, max_zoom] then apply to all spatial dims to keep the original spatial shape ratio. If a sequence, max_zoom should contain one value for each spatial axis. If 2 values provided for 3D data, use the first value for both H & W dims to keep the same zoom ratio.
- prob: Probability of zooming.
- whole_batch: Flag to apply transform on whole batch.
If False, each image in the batch is randomly transformed It True, entire batch is transformed randomly.
- Returns
- outcupy.ndarray or numpy.ndarray
Output data. Same dimensions and type as input.
- Raises
- TypeError
If input ‘img’ is not cupy.ndarray or numpy.ndarray
Examples
>>> import cucim.core.operations.intensity as its >>> # input is channel first 3d array >>> output_array = its.rand_zoom(input_arr)
- cucim.core.operations.intensity.scale_intensity_range(img: Any, b_max: float, b_min: float, a_max: float, a_min: float, clip: bool = False) Any#
Apply intensity scaling to the input array. Scaling from [a_min, a_max] to [b_min, b_max] with clip option.
- Parameters
- imgchannel first, cupy.ndarray or numpy.ndarray
Input data of shape (C, H, W). Can also batch process input of shape (N, C, H, W). Can be a numpy.ndarray or cupy.ndarray.
- b_minfloat
intensity target range min.
- b_maxfloat
intensity target range max.
- a_minfloat
intensity original range min.
- a_maxfloat
intensity original range max.
- clipfloat
whether to perform clip after scaling.
- Returns
- outcupy.ndarray or numpy.ndarray
Output data. Same dimensions and type as input.
- Raises
- TypeError
If input ‘img’ is not cupy.ndarray or numpy.ndarray
- ValueError
If input original intensity min and max are same
Examples
>>> import cucim.core.operations.intensity as its >>> # input is channel first 3d array >>> output_array = its.scale_intensity_range(input_arr, 0.0, 255.0, -1.0, 1.0, False)
- cucim.core.operations.intensity.zoom(img: Any, zoom_factor: Sequence[float])#
Zooms an ND image
- Parameters
- imgchannel first, cupy.ndarray or numpy.ndarray
Input data of shape (C, H, W). Can also batch process input of shape (N, C, H, W). Can be a numpy.ndarray or cupy.ndarray.
- zoom_factor: Sequence[float]
The zoom factor along the spatial axes. Zoom factor should contain one value for each spatial axis.
- Returns
- ——-
- outcupy.ndarray or numpy.ndarray
Output data. Same dimensions and type as input.
- Raises
- TypeError
If input ‘img’ is not cupy.ndarray or numpy.ndarray
Examples
>>> import cucim.core.operations.intensity as its >>> # input is channel first 3d array >>> output_array = its.zoom(input_arr,[1.1,1.1])
morphology#
- cucim.core.operations.morphology.distance_transform_edt(image, sampling=None, return_distances=True, return_indices=False, distances=None, indices=None, *, block_params=None, float64_distances=False)#
Exact Euclidean distance transform.
This function calculates the distance transform of the input, by replacing each foreground (non-zero) element, with its shortest distance to the background (any zero-valued element).
In addition to the distance transform, the feature transform can be calculated. In this case the index of the closest background element to each foreground element is returned in a separate array.
- Parameters
- imagearray_like
Input data to transform. Can be any type but will be converted into binary: 1 wherever image equates to True, 0 elsewhere.
- samplingfloat, or sequence of float, optional
Spacing of elements along each dimension. If a sequence, must be of length equal to the image rank; if a single number, this is used for all axes. If not specified, a grid spacing of unity is implied.
- return_distancesbool, optional
Whether to calculate the distance transform.
- return_indicesbool, optional
Whether to calculate the feature transform.
- distancesfloat32 cupy.ndarray, optional
An output array to store the calculated distance transform, instead of returning it. return_distances must be True. It must be the same shape as image.
- indicesint32 cupy.ndarray, optional
An output array to store the calculated feature transform, instead of returning it. return_indicies must be True. Its shape must be (image.ndim,) + image.shape.
- Returns
- distancesfloat64 ndarray, optional
The calculated distance transform. Returned only when return_distances is True and distances is not supplied. It will have the same shape as image.
- indicesint32 ndarray, optional
The calculated feature transform. It has an image-shaped array for each dimension of the image. See example below. Returned only when return_indices is True and indices is not supplied.
- Other Parameters
- block_params3-tuple of int
The m1, m2, m3 algorithm parameters as described in [2]. If None, suitable defaults will be chosen. Note: This parameter is specific to cuCIM and does not exist in SciPy.
- float64_distancesbool, optional
If True, use double precision in the distance computation (to match SciPy behavior). Otherwise, single precision will be used for efficiency. Note: This parameter is specific to cuCIM and does not exist in SciPy.
Notes
The Euclidean distance transform gives values of the Euclidean distance.
\[y_i = \sqrt{\sum_{i}^{n} (x[i] - b[i])^2}\]where \(b[i]\) is the background point (value 0) with the smallest Euclidean distance to input points \(x[i]\), and \(n\) is the number of dimensions.
Note that the indices output may differ from the one given by
scipy.ndimage.distance_transform_edt()in the case of input pixels that are equidistant from multiple background points.The parallel banding algorithm implemented here was originally described in [1]. The kernels used here correspond to the revised PBA+ implementation that is described on the author’s website [2]. The source code of the author’s PBA+ implementation is available at [3].
References
- 1
Thanh-Tung Cao, Ke Tang, Anis Mohamed, and Tiow-Seng Tan. 2010. Parallel Banding Algorithm to compute exact distance transform with the GPU. In Proceedings of the 2010 ACM SIGGRAPH symposium on Interactive 3D Graphics and Games (I3D ’10). Association for Computing Machinery, New York, NY, USA, 83–90. DOI:https://doi.org/10.1145/1730804.1730818
- 2(1,2)
- 3
Examples
>>> import cupy as cp >>> from cucim.core.operations import morphology >>> a = cp.array(([0,1,1,1,1], ... [0,0,1,1,1], ... [0,1,1,1,1], ... [0,1,1,1,0], ... [0,1,1,0,0])) >>> morphology.distance_transform_edt(a) array([[ 0. , 1. , 1.4142, 2.2361, 3. ], [ 0. , 0. , 1. , 2. , 2. ], [ 0. , 1. , 1.4142, 1.4142, 1. ], [ 0. , 1. , 1.4142, 1. , 0. ], [ 0. , 1. , 1. , 0. , 0. ]])
With a sampling of 2 units along x, 1 along y:
>>> morphology.distance_transform_edt(a, sampling=[2,1]) array([[ 0. , 1. , 2. , 2.8284, 3.6056], [ 0. , 0. , 1. , 2. , 3. ], [ 0. , 1. , 2. , 2.2361, 2. ], [ 0. , 1. , 2. , 1. , 0. ], [ 0. , 1. , 1. , 0. , 0. ]])
Asking for indices as well:
>>> edt, inds = morphology.distance_transform_edt(a, return_indices=True) >>> inds array([[[0, 0, 1, 1, 3], [1, 1, 1, 1, 3], [2, 2, 1, 3, 3], [3, 3, 4, 4, 3], [4, 4, 4, 4, 4]], [[0, 0, 1, 1, 4], [0, 1, 1, 1, 4], [0, 0, 1, 4, 4], [0, 0, 3, 3, 4], [0, 0, 3, 3, 4]]])
spatial#
- cucim.core.operations.spatial.image_flip(img: Any, spatial_axis: ()) Any#
Shape preserving order reversal of elements in input array along the given spatial axis
- Parameters
- imgcupy.ndarray or numpy.ndarray
Input data. Can be numpy.ndarray or cupy.ndarray
- spatial_axistuple
spatial axis along which to flip over the input array
- Returns
- outcupy.ndarray or numpy.ndarray
Output data. Same dimensions and type as input.
- Raises
- TypeError
If input ‘img’ is not cupy.ndarray or numpy.ndarray
Examples
>>> import cucim.core.operations.spatial as spt >>> # input is channel first 3d array >>> output_array = spt.image_flip(input_arr, (1, 2))
- cucim.core.operations.spatial.image_rotate_90(img: Any, k: int, spatial_axis: ()) Any#
Rotate input array by 90 degress along the given axis
- Parameters
- imgcupy.ndarray or numpy.ndarray
Input data. Can be numpy.ndarray or cupy.ndarray
- kint
number of times to rotate
- spatial_axistuple
spatial axis along which to rotate the input array by 90 degrees
- Returns
- outcupy.ndarray or numpy.ndarray
Output data. Same dimensions and type as input.
- Raises
- TypeError
If input ‘img’ is not cupy.ndarray or numpy.ndarray
Examples
>>> import cucim.core.operations.spatial as spt >>> # input is channel first 3d array >>> output_array = spt.image_rotate_90(input_arr,1,(1,2))
- cucim.core.operations.spatial.rand_image_flip(img: Any, spatial_axis: (), prob: float = 0.1, whole_batch: bool = False) Any#
Randomly flips the image along axis.
- Parameters
- imgcupy.ndarray or numpy.ndarray
Input data. Can be numpy.ndarray or cupy.ndarray
- prob: Probability of flipping.
- spatial_axistuple
spatial axis along which to flip over the input array
- whole_batch: Flag to apply transform on whole batch.
If False, each image in the batch is randomly transformed It True, entire batch is transformed randomly.
- Returns
- outcupy.ndarray or numpy.ndarray
Output data. Same dimensions and type as input.
- Raises
- TypeError
If input ‘img’ is not cupy.ndarray or numpy.ndarray
Examples
>>> import cucim.core.operations.spatial as spt >>> # input is channel first 3d array >>> output_array = spt.rand_image_flip(input_arr,spatial_axis=(1,2))
- cucim.core.operations.spatial.rand_image_rotate_90(img: Any, spatial_axis: (), prob: float = 0.1, max_k: int = 3, whole_batch: bool = False) Any#
With probability prob, input arrays are rotated by 90 degrees in the plane specified by spatial_axis.
- Parameters
- imgcupy.ndarray or numpy.ndarray
Input data. Can be numpy.ndarray or cupy.ndarray
- prob: probability of rotating.
(Default 0.1, with 10% probability it returns a rotated array)
- max_k: number of rotations
will be sampled from np.random.randint(max_k) + 1, (Default 3).
- spatial_axistuple
spatial axis along which to rotate the input array by 90 degrees
- whole_batch: Flag to apply transform on whole batch.
If False, each image in the batch is randomly transformed It True, entire batch is transformed randomly.
- Returns
- outcupy.ndarray or numpy.ndarray
Output data. Same dimensions and type as input.
- Raises
- TypeError
If input ‘img’ is not cupy.ndarray or numpy.ndarray
Examples
>>> import cucim.core.operations.spatial as spt >>> # input is channel first 3d array >>> output_array = spt.rand_image_rotate_90(input_arr, spatial_axis=(1, 2))
skimage Submodules#
color#
- cucim.skimage.color.combine_stains(stains, conv_matrix, *, channel_axis=-1)#
Stain to RGB color space conversion.
- Parameters
- stains(…, 3, …) array_like
The image in stain color space. By default, the final dimension denotes channels.
- conv_matrix: ndarray
The stain separation matrix as described by G. Landini [1].
- channel_axisint, optional
This parameter indicates which axis of the array corresponds to channels.
- Returns
- out(…, 3, …) ndarray
The image in RGB format. Same dimensions as input.
- Raises
- ValueError
If stains is not at least 2-D with shape (…, 3, …).
Notes
Stain combination matrices available in the
colormodule and their respective colorspace:rgb_from_hed: Hematoxylin + Eosin + DABrgb_from_hdx: Hematoxylin + DABrgb_from_fgx: Feulgen + Light Greenrgb_from_bex: Giemsa stain : Methyl Blue + Eosinrgb_from_rbd: FastRed + FastBlue + DABrgb_from_gdx: Methyl Green + DABrgb_from_hax: Hematoxylin + AECrgb_from_bro: Blue matrix Anilline Blue + Red matrix Azocarmine + Orange matrix Orange-Grgb_from_bpx: Methyl Blue + Ponceau Fuchsinrgb_from_ahx: Alcian Blue + Hematoxylinrgb_from_hpx: Hematoxylin + PAS
References
- 1
- 2
A. C. Ruifrok and D. A. Johnston, “Quantification of histochemical staining by color deconvolution,” Anal. Quant. Cytol. Histol., vol. 23, no. 4, pp. 291–299, Aug. 2001.
Examples
>>> import cupy as cp >>> from skimage import data >>> from cucim.skimage.color import (separate_stains, combine_stains, ... hdx_from_rgb, rgb_from_hdx) >>> ihc = cp.array(data.immunohistochemistry()) >>> ihc_hdx = separate_stains(ihc, hdx_from_rgb) >>> ihc_rgb = combine_stains(ihc_hdx, rgb_from_hdx)
- cucim.skimage.color.convert_colorspace(arr, fromspace, tospace, *, channel_axis=-1)#
Convert an image array to a new color space.
- Valid color spaces are:
‘RGB’, ‘HSV’, ‘RGB CIE’, ‘XYZ’, ‘YUV’, ‘YIQ’, ‘YPbPr’, ‘YCbCr’, ‘YDbDr’
- Parameters
- arr(…, 3, …) array_like
The image to convert. By default, the final dimension denotes channels.
- fromspacestr
The color space to convert from. Can be specified in lower case.
- tospacestr
The color space to convert to. Can be specified in lower case.
- channel_axisint, optional
This parameter indicates which axis of the array corresponds to channels.
- Returns
- out(…, 3, …) ndarray
The converted image. Same dimensions as input.
- Raises
- ValueError
If fromspace is not a valid color space
- ValueError
If tospace is not a valid color space
Notes
Conversion is performed through the “central” RGB color space, i.e. conversion from XYZ to HSV is implemented as
XYZ -> RGB -> HSVinstead of directly.Examples
>>> import cupy as cp >>> from skimage import data >>> img = cp.array(data.astronaut()) >>> img_hsv = convert_colorspace(img, 'RGB', 'HSV')
- cucim.skimage.color.deltaE_cie76(lab1, lab2, channel_axis=-1)#
Euclidean distance between two points in Lab color space
- Parameters
- lab1array_like
reference color (Lab colorspace)
- lab2array_like
comparison color (Lab colorspace)
- channel_axisint, optional
This parameter indicates which axis of the arrays corresponds to channels.
- Returns
- dEarray_like
distance between colors lab1 and lab2
References
- 1
- 2
A. R. Robertson, “The CIE 1976 color-difference formulae,” Color Res. Appl. 2, 7-11 (1977).
- cucim.skimage.color.deltaE_ciede2000(lab1, lab2, kL=1, kC=1, kH=1, *, channel_axis=-1)#
Color difference as given by the CIEDE 2000 standard.
CIEDE 2000 is a major revision of CIDE94. The perceptual calibration is largely based on experience with automotive paint on smooth surfaces.
- Parameters
- lab1array_like
reference color (Lab colorspace)
- lab2array_like
comparison color (Lab colorspace)
- kLfloat (range), optional
lightness scale factor, 1 for “acceptably close”; 2 for “imperceptible” see deltaE_cmc
- kCfloat (range), optional
chroma scale factor, usually 1
- kHfloat (range), optional
hue scale factor, usually 1
- channel_axisint, optional
This parameter indicates which axis of the arrays corresponds to channels.
- Returns
- deltaEarray_like
The distance between lab1 and lab2
Notes
CIEDE 2000 assumes parametric weighting factors for the lightness, chroma, and hue (kL, kC, kH respectively). These default to 1.
References
- 1
- 2
http://www.ece.rochester.edu/~gsharma/ciede2000/ciede2000noteCRNA.pdf DOI:10.1364/AO.33.008069
- 3
M. Melgosa, J. Quesada, and E. Hita, “Uniformity of some recent color metrics tested with an accurate color-difference tolerance dataset,” Appl. Opt. 33, 8069-8077 (1994).
- cucim.skimage.color.deltaE_ciede94(lab1, lab2, kH=1, kC=1, kL=1, k1=0.045, k2=0.015, *, channel_axis=-1)#
Color difference according to CIEDE 94 standard
Accommodates perceptual non-uniformities through the use of application specific scale factors (kH, kC, kL, k1, and k2).
- Parameters
- lab1array_like
reference color (Lab colorspace)
- lab2array_like
comparison color (Lab colorspace)
- kHfloat, optional
Hue scale
- kCfloat, optional
Chroma scale
- kLfloat, optional
Lightness scale
- k1float, optional
first scale parameter
- k2float, optional
second scale parameter
- channel_axisint, optional
This parameter indicates which axis of the arrays corresponds to channels.
- Returns
- dEarray_like
color difference between lab1 and lab2
Notes
deltaE_ciede94 is not symmetric with respect to lab1 and lab2. CIEDE94 defines the scales for the lightness, hue, and chroma in terms of the first color. Consequently, the first color should be regarded as the “reference” color.
kL, k1, k2 depend on the application and default to the values suggested for graphic arts
Parameter
Graphic Arts
Textiles
kL
1.000
2.000
k1
0.045
0.048
k2
0.015
0.014
References
- cucim.skimage.color.deltaE_cmc(lab1, lab2, kL=1, kC=1, *, channel_axis=-1)#
Color difference from the CMC l:c standard.
This color difference was developed by the Colour Measurement Committee (CMC) of the Society of Dyers and Colourists (United Kingdom). It is intended for use in the textile industry.
The scale factors kL, kC set the weight given to differences in lightness and chroma relative to differences in hue. The usual values are
kL=2,kC=1for “acceptability” andkL=1,kC=1for “imperceptibility”. Colors withdE > 1are “different” for the given scale factors.- Parameters
- lab1array_like
reference color (Lab colorspace)
- lab2array_like
comparison color (Lab colorspace)
- channel_axisint, optional
This parameter indicates which axis of the arrays corresponds to channels.
- Returns
- dEarray_like
distance between colors lab1 and lab2
Notes
deltaE_cmc the defines the scales for the lightness, hue, and chroma in terms of the first color. Consequently
deltaE_cmc(lab1, lab2) != deltaE_cmc(lab2, lab1)References
- 1
- 2
http://www.brucelindbloom.com/index.html?Eqn_DeltaE_CIE94.html
- 3
F. J. J. Clarke, R. McDonald, and B. Rigg, “Modification to the JPC79 colour-difference formula,” J. Soc. Dyers Colour. 100, 128-132 (1984).
- cucim.skimage.color.gray2rgb(image, *, channel_axis=-1)#
Create an RGB representation of a gray-level image.
- Parameters
- imagearray_like
Input image.
- channel_axisint, optional
This parameter indicates which axis of the output array will correspond to channels.
- Returns
- rgb(…, 3, …) ndarray
RGB image. A new dimension of length 3 is added to input image.
Notes
If the input is a 1-dimensional image of shape
(M, ), the output will be shape(M, 3).
- cucim.skimage.color.gray2rgba(image, alpha=None, *, channel_axis=-1)#
Create a RGBA representation of a gray-level image.
- Parameters
- imagearray_like
Input image.
- alphaarray_like, optional
Alpha channel of the output image. It may be a scalar or an array that can be broadcast to
image. If not specified it is set to the maximum limit corresponding to theimagedtype.- channel_axisint, optional
This parameter indicates which axis of the output array will correspond to channels.
- Returns
- rgbandarray
RGBA image. A new dimension of length 4 is added to input image shape.
- cucim.skimage.color.hed2rgb(hed, *, channel_axis=-1)#
Haematoxylin-Eosin-DAB (HED) to RGB color space conversion.
- Parameters
- hed(…, 3, …) array_like
The image in the HED color space. By default, the final dimension denotes channels.
- channel_axisint, optional
This parameter indicates which axis of the array corresponds to channels.
- Returns
- out(…, 3, …) ndarray
The image in RGB. Same dimensions as input.
- Raises
- ValueError
If hed is not at least 2-D with shape (…, 3, …).
References
- 1
A. C. Ruifrok and D. A. Johnston, “Quantification of histochemical staining by color deconvolution.,” Analytical and quantitative cytology and histology / the International Academy of Cytology [and] American Society of Cytology, vol. 23, no. 4, pp. 291-9, Aug. 2001.
Examples
>>> import cupy as cp >>> from skimage import data >>> from cucim.skimage.color import rgb2hed, hed2rgb >>> ihc = cp.array(data.immunohistochemistry()) >>> ihc_hed = rgb2hed(ihc) >>> ihc_rgb = hed2rgb(ihc_hed)
- cucim.skimage.color.hsv2rgb(hsv, *, channel_axis=-1)#
HSV to RGB color space conversion.
- Parameters
- hsv(…, 3, …) array_like
The image in HSV format. By default, the final dimension denotes channels.
- channel_axisint, optional
This parameter indicates which axis of the array corresponds to channels.
- Returns
- out(…, 3, …) ndarray
The image in RGB format. Same dimensions as input.
- Raises
- ValueError
If hsv is not at least 2-D with shape (…, 3, …).
Notes
Conversion between RGB and HSV color spaces results in some loss of precision, due to integer arithmetic and rounding [1].
References
Examples
>>> import cupy as cp >>> from skimage import data >>> img = cp.array(data.astronaut()) >>> img_hsv = rgb2hsv(img) >>> img_rgb = hsv2rgb(img_hsv)
- cucim.skimage.color.lab2lch(lab, *, channel_axis=-1)#
CIE-LAB to CIE-LCH color space conversion.
LCH is the cylindrical representation of the LAB (Cartesian) colorspace
- Parameters
- lab(…, 3, …) array_like
The N-D image in CIE-LAB format. The last (
N+1-th) dimension must have at least 3 elements, corresponding to theL,a, andbcolor channels. Subsequent elements are copied.- channel_axisint, optional
This parameter indicates which axis of the array corresponds to channels.
- Returns
- out(…, 3, …) ndarray
The image in LCH format, in a N-D array with same shape as input lab.
- Raises
- ValueError
If lch does not have at least 3 color channels (i.e. l, a, b).
Notes
The Hue is expressed as an angle between
(0, 2*pi)Examples
>>> from skimage import data >>> from cucim.skimage.color import rgb2lab, lab2lch >>> img = cp.array(data.astronaut()) >>> img_lab = rgb2lab(img) >>> img_lch = lab2lch(img_lab)
- cucim.skimage.color.lab2rgb(lab, illuminant='D65', observer='2', *, channel_axis=-1)#
Lab to RGB color space conversion.
- Parameters
- lab(…, 3, …) array_like
The image in Lab format. By default, the final dimension denotes channels.
- illuminant{“A”, “B”, “C”, “D50”, “D55”, “D65”, “D75”, “E”}, optional
The name of the illuminant (the function is NOT case sensitive).
- observer{“2”, “10”, “R”}, optional
The aperture angle of the observer.
- channel_axisint, optional
This parameter indicates which axis of the array corresponds to channels.
- Returns
- out(…, 3, …) ndarray
The image in RGB format. Same dimensions as input.
- Raises
- ValueError
If lab is not at least 2-D with shape (…, 3, …).
Notes
This function uses lab2xyz and xyz2rgb. By default Observer=”2”, Illuminant=”D65”. CIE XYZ tristimulus values x_ref=95.047, y_ref=100., z_ref=108.883. See function get_xyz_coords for a list of supported illuminants.
References
- cucim.skimage.color.lab2xyz(lab, illuminant='D65', observer='2', *, channel_axis=-1)#
CIE-LAB to XYZcolor space conversion.
- Parameters
- lab(…, 3, …) array_like
The image in Lab format. By default, the final dimension denotes channels.
- illuminant{“A”, “B”, “C”, “D50”, “D55”, “D65”, “D75”, “E”}, optional
The name of the illuminant (the function is NOT case sensitive).
- observer{“2”, “10”, “R”}, optional
The aperture angle of the observer.
- channel_axisint, optional
This parameter indicates which axis of the array corresponds to channels.
- Returns
- out(…, 3, …) ndarray
The image in XYZ format. Same dimensions as input.
- Raises
- ValueError
If lab is not at least 2-D with shape (…, 3, …).
- ValueError
If either the illuminant or the observer angle are not supported or unknown.
- UserWarning
If any of the pixels are invalid (Z < 0).
Notes
By default Observer=”2”, Illuminant=”D65”. CIE XYZ tristimulus values x_ref = 95.047, y_ref = 100., z_ref = 108.883. See function ‘get_xyz_coords’ for a list of supported illuminants.
References
- cucim.skimage.color.label2rgb(label, image=None, colors=None, alpha=0.3, bg_label=0, bg_color=(0, 0, 0), image_alpha=1, kind='overlay', *, saturation=0, channel_axis=-1)#
Return an RGB image where color-coded labels are painted over the image.
- Parameters
- labelndarray
Integer array of labels with the same shape as image.
- imagendarray, optional
Image used as underlay for labels. It should have the same shape as labels, optionally with an additional RGB (channels) axis. If image is an RGB image, it is converted to grayscale before coloring.
- colorslist, optional
List of colors. If the number of labels exceeds the number of colors, then the colors are cycled.
- alphafloat [0, 1], optional
Opacity of colorized labels. Ignored if image is None.
- bg_labelint, optional
Label that’s treated as the background. If bg_label is specified, bg_color is None, and kind is overlay, background is not painted by any colors.
- bg_colorstr or array, optional
Background color. Must be a name in color_dict or RGB float values between [0, 1].
- image_alphafloat [0, 1], optional
Opacity of the image.
- kindstring, one of {‘overlay’, ‘avg’}
The kind of color image desired. ‘overlay’ cycles over defined colors and overlays the colored labels over the original image. ‘avg’ replaces each labeled segment with its average color, for a stained-class or pastel painting appearance.
- saturationfloat [0, 1], optional
Parameter to control the saturation applied to the original image between fully saturated (original RGB, saturation=1) and fully unsaturated (grayscale, saturation=0). Only applies when kind=’overlay’.
- channel_axisint, optional
This parameter indicates which axis of the output array will correspond to channels. If image is provided, this must also match the axis of image that corresponds to channels.
- Returns
- resultarray of float, shape (M, N, 3)
The result of blending a cycling colormap (colors) for each distinct value in label with the image, at a certain alpha value.
- cucim.skimage.color.lch2lab(lch, *, channel_axis=-1)#
CIE-LCH to CIE-LAB color space conversion.
LCH is the cylindrical representation of the LAB (Cartesian) colorspace
- Parameters
- lch(…, 3, …) array_like
The N-D image in CIE-LCH format. The last (
N+1-th) dimension must have at least 3 elements, corresponding to theL,a, andbcolor channels. Subsequent elements are copied.- channel_axisint, optional
This parameter indicates which axis of the array corresponds to channels.
- Returns
- out(…, 3, …) ndarray
The image in LAB format, with same shape as input lch.
- Raises
- ValueError
If lch does not have at least 3 color channels (i.e. l, c, h).
Examples
>>> from skimage import data >>> from cucim.skimage.color import rgb2lab, lch2lab >>> img = cp.array(data.astronaut()) >>> img_lab = rgb2lab(img) >>> img_lch = lab2lch(img_lab) >>> img_lab2 = lch2lab(img_lch)
- cucim.skimage.color.luv2rgb(luv, *, channel_axis=-1)#
Luv to RGB color space conversion.
- Parameters
- luv(…, 3, …) array_like
The image in CIE Luv format. By default, the final dimension denotes channels.
- Returns
- out(…, 3, …) ndarray
The image in RGB format. Same dimensions as input.
- Raises
- ValueError
If luv is not at least 2-D with shape (…, 3, …).
Notes
This function uses luv2xyz and xyz2rgb.
- cucim.skimage.color.luv2xyz(luv, illuminant='D65', observer='2', *, channel_axis=-1)#
CIE-Luv to XYZ color space conversion.
- Parameters
- luv(…, 3, …) array_like
The image in CIE-Luv format. By default, the final dimension denotes channels.
- illuminant{“A”, “B”, “C”, “D50”, “D55”, “D65”, “D75”, “E”}, optional
The name of the illuminant (the function is NOT case sensitive).
- observer{“2”, “10”, “R”}, optional
The aperture angle of the observer.
- channel_axisint, optional
This parameter indicates which axis of the array corresponds to channels.
- Returns
- out(…, 3, …) ndarray
The image in XYZ format. Same dimensions as input.
- Raises
- ValueError
If luv is not at least 2-D with shape (…, 3, …).
- ValueError
If either the illuminant or the observer angle are not supported or unknown.
Notes
XYZ conversion weights use observer=2A. Reference whitepoint for D65 Illuminant, with XYZ tristimulus values of
(95.047, 100., 108.883). See function ‘get_xyz_coords’ for a list of supported illuminants.References
- cucim.skimage.color.rgb2gray(rgb, *, channel_axis=-1)#
Compute luminance of an RGB image.
- Parameters
- rgb(…, 3, …) array_like
The image in RGB format. By default, the final dimension denotes channels.
- Returns
- outndarray
The luminance image - an array which is the same size as the input array, but with the channel dimension removed.
- Raises
- ValueError
If rgb is not at least 2-D with shape (…, 3, …).
Notes
The weights used in this conversion are calibrated for contemporary CRT phosphors:
Y = 0.2125 R + 0.7154 G + 0.0721 B
If there is an alpha channel present, it is ignored.
References
Examples
>>> import cupy as cp >>> from cucim.skimage.color import rgb2gray >>> from skimage import data >>> img = cp.array(data.astronaut()) >>> img_gray = rgb2gray(img)
- cucim.skimage.color.rgb2hed(rgb, *, channel_axis=-1)#
RGB to Haematoxylin-Eosin-DAB (HED) color space conversion.
- Parameters
- rgb(…, 3, …) array_like
The image in RGB format. By default, the final dimension denotes channels.
- channel_axisint, optional
This parameter indicates which axis of the array corresponds to channels.
- Returns
- out(…, 3, …) ndarray
The image in HED format. Same dimensions as input.
- Raises
- ValueError
If rgb is not at least 2-D with shape (…, 3, …).
References
- 1
A. C. Ruifrok and D. A. Johnston, “Quantification of histochemical staining by color deconvolution.,” Analytical and quantitative cytology and histology / the International Academy of Cytology [and] American Society of Cytology, vol. 23, no. 4, pp. 291-9, Aug. 2001.
Examples
>>> import cupy as cp >>> from skimage import data >>> from cucim.skimage.color import rgb2hed >>> ihc = cp.array(data.immunohistochemistry()) >>> ihc_hed = rgb2hed(ihc)
- cucim.skimage.color.rgb2hsv(rgb, *, channel_axis=-1)#
RGB to HSV color space conversion.
- Parameters
- rgb(…, 3, …) array_like
The image in RGB format. By default, the final dimension denotes channels.
- channel_axisint, optional
This parameter indicates which axis of the array corresponds to channels.
- Returns
- out(…, 3, …) ndarray
The image in HSV format. Same dimensions as input.
- Raises
- ValueError
If rgb is not at least 2-D with shape (…, 3, …).
Notes
Conversion between RGB and HSV color spaces results in some loss of precision, due to integer arithmetic and rounding [1].
References
Examples
>>> import cupy as cp >>> from cucim.skimage import color >>> from skimage import data >>> img = cp.array(data.astronaut()) >>> img_hsv = color.rgb2hsv(img)
- cucim.skimage.color.rgb2lab(rgb, illuminant='D65', observer='2', *, channel_axis=-1)#
Conversion from the sRGB color space (IEC 61966-2-1:1999) to the CIE Lab colorspace under the given illuminant and observer.
- Parameters
- rgb(…, 3, …) array_like
The image in RGB format. By default, the final dimension denotes channels.
- illuminant{“A”, “B”, “C”, “D50”, “D55”, “D65”, “D75”, “E”}, optional
The name of the illuminant (the function is NOT case sensitive).
- observer{“2”, “10”, “R”}, optional
The aperture angle of the observer.
- channel_axisint, optional
This parameter indicates which axis of the array corresponds to channels.
- Returns
- out(…, 3, …) ndarray
The image in Lab format. Same dimensions as input.
- Raises
- ValueError
If rgb is not at least 2-D with shape (…, 3, …).
Notes
RGB is a device-dependent color space so, if you use this function, be sure that the image you are analyzing has been mapped to the sRGB color space.
This function uses rgb2xyz and xyz2lab. By default Observer=”2”, Illuminant=”D65”. CIE XYZ tristimulus values x_ref=95.047, y_ref=100., z_ref=108.883. See function get_xyz_coords for a list of supported illuminants.
References
- cucim.skimage.color.rgb2luv(rgb, *, channel_axis=-1)#
RGB to CIE-Luv color space conversion.
- Parameters
- rgb(…, 3, …) array_like
The image in RGB format. By default, the final dimension denotes channels.
- channel_axisint, optional
This parameter indicates which axis of the array corresponds to channels.
- Returns
- out(…, 3, …) ndarray
The image in CIE Luv format. Same dimensions as input.
- Raises
- ValueError
If rgb is not at least 2-D with shape (…, 3, …).
Notes
This function uses rgb2xyz and xyz2luv.
References
- cucim.skimage.color.rgb2rgbcie(rgb, *, channel_axis=-1)#
RGB to RGB CIE color space conversion.
- Parameters
- rgb(…, 3, …) array_like
The image in RGB format. By default, the final dimension denotes channels.
- channel_axisint, optional
This parameter indicates which axis of the array corresponds to channels.
- Returns
- out(…, 3, …) ndarray
The image in RGB CIE format. Same dimensions as input.
- Raises
- ValueError
If rgb is not at least 2-D with shape (…, 3, …).
References
Examples
>>> from skimage import data >>> from cucim.skimage.color import rgb2rgbcie >>> img = cp.array(data.astronaut()) >>> img_rgbcie = rgb2rgbcie(img)
- cucim.skimage.color.rgb2xyz(rgb, *, channel_axis=-1)#
RGB to XYZ color space conversion.
- Parameters
- rgb(…, 3, …) array_like
The image in RGB format. By default, the final dimension denotes channels.
- channel_axisint, optional
This parameter indicates which axis of the array corresponds to channels.
- Returns
- out(…, 3, …) ndarray
The image in XYZ format. Same dimensions as input.
- Raises
- ValueError
If rgb is not at least 2-D with shape (…, 3, …).
Notes
The CIE XYZ color space is derived from the CIE RGB color space. Note however that this function converts from sRGB.
References
Examples
>>> import cupy as cp >>> from skimage import data >>> img = cp.array(data.astronaut()) >>> img_xyz = rgb2xyz(img)
- cucim.skimage.color.rgb2ycbcr(rgb, *, channel_axis=-1)#
RGB to YCbCr color space conversion.
- Parameters
- rgb(…, 3, …) array_like
The image in RGB format. By default, the final dimension denotes channels.
- channel_axisint, optional
This parameter indicates which axis of the array corresponds to channels.
- Returns
- out(…, 3, …) ndarray
The image in YCbCr format. Same dimensions as input.
- Raises
- ValueError
If rgb is not at least 2-D with shape (…, 3, …).
Notes
Y is between 16 and 235. This is the color space commonly used by video codecs; it is sometimes incorrectly called “YUV”.
References
- cucim.skimage.color.rgb2ydbdr(rgb, *, channel_axis=-1)#
RGB to YDbDr color space conversion.
- Parameters
- rgb(…, 3, …) array_like
The image in RGB format. By default, the final dimension denotes channels.
- channel_axisint, optional
This parameter indicates which axis of the array corresponds to channels.
- Returns
- out(…, 3, …) ndarray
The image in YDbDr format. Same dimensions as input.
- Raises
- ValueError
If rgb is not at least 2-D with shape (…, 3, …).
Notes
This is the color space commonly used by video codecs. It is also the reversible color transform in JPEG2000.
References
- cucim.skimage.color.rgb2yiq(rgb, *, channel_axis=-1)#
RGB to YIQ color space conversion.
- Parameters
- rgb(…, 3, …) array_like
The image in RGB format. By default, the final dimension denotes channels.
- channel_axisint, optional
This parameter indicates which axis of the array corresponds to channels.
- Returns
- out(…, 3, …) ndarray
The image in YIQ format. Same dimensions as input.
- Raises
- ValueError
If rgb is not at least 2-D with shape (…, 3, …).
- cucim.skimage.color.rgb2ypbpr(rgb, *, channel_axis=-1)#
RGB to YPbPr color space conversion.
- Parameters
- rgb(…, 3, …) array_like
The image in RGB format. By default, the final dimension denotes channels.
- channel_axisint, optional
This parameter indicates which axis of the array corresponds to channels.
- Returns
- out(…, 3, …) ndarray
The image in YPbPr format. Same dimensions as input.
- Raises
- ValueError
If rgb is not at least 2-D with shape (…, 3, …).
References
- cucim.skimage.color.rgb2yuv(rgb, *, channel_axis=-1)#
RGB to YUV color space conversion.
- Parameters
- rgb(…, 3, …) array_like
The image in RGB format. By default, the final dimension denotes channels.
- channel_axisint, optional
This parameter indicates which axis of the array corresponds to channels.
- Returns
- out(…, 3, …) ndarray
The image in YUV format. Same dimensions as input.
- Raises
- ValueError
If rgb is not at least 2-D with shape (…, 3, …).
Notes
Y is between 0 and 1. Use YCbCr instead of YUV for the color space commonly used by video codecs, where Y ranges from 16 to 235.
References
- cucim.skimage.color.rgba2rgb(rgba, background=(1, 1, 1), *, channel_axis=-1)#
RGBA to RGB conversion using alpha blending [1].
- Parameters
- rgba(…, 4, …) array_like
The image in RGBA format. By default, the final dimension denotes channels.
- backgroundarray_like
The color of the background to blend the image with (3 floats between 0 to 1 - the RGB value of the background).
- channel_axisint, optional
This parameter indicates which axis of the array corresponds to channels.
- Returns
- out(…, 3, …) ndarray
The image in RGB format. Same dimensions as input.
- Raises
- ValueError
If rgba is not at least 2D with shape (…, 4, …).
References
Examples
>>> import cupy as cp >>> from cucim.skimage import color >>> from skimage import data >>> img_rgba = cp.array(data.logo()) >>> img_rgb = color.rgba2rgb(img_rgba)
- cucim.skimage.color.rgbcie2rgb(rgbcie, *, channel_axis=-1)#
RGB CIE to RGB color space conversion.
- Parameters
- rgbcie(…, 3, …) array_like
The image in RGB CIE format. By default, the final dimension denotes channels.
- channel_axisint, optional
This parameter indicates which axis of the array corresponds to channels.
- Returns
- out(…, 3, …) ndarray
The image in RGB format. Same dimensions as input.
- Raises
- ValueError
If rgbcie is not at least 2-D with shape (…, 3, …).
References
Examples
>>> from skimage import data >>> from cucim.skimage.color import rgb2rgbcie, rgbcie2rgb >>> img = cp.array(data.astronaut()) >>> img_rgbcie = rgb2rgbcie(img) >>> img_rgb = rgbcie2rgb(img_rgbcie)
- cucim.skimage.color.separate_stains(rgb, conv_matrix, *, channel_axis=-1)#
RGB to stain color space conversion.
- Parameters
- rgb(…, 3, …) array_like
The image in RGB format. By default, the final dimension denotes channels.
- conv_matrix: ndarray
The stain separation matrix as described by G. Landini [1].
- channel_axisint, optional
This parameter indicates which axis of the array corresponds to channels.
- Returns
- out(…, 3, …) ndarray
The image in stain color space. Same dimensions as input.
- Raises
- ValueError
If rgb is not at least 2-D with shape (…, 3, …).
Notes
Stain separation matrices available in the
colormodule and their respective colorspace:hed_from_rgb: Hematoxylin + Eosin + DABhdx_from_rgb: Hematoxylin + DABfgx_from_rgb: Feulgen + Light Greenbex_from_rgb: Giemsa stain : Methyl Blue + Eosinrbd_from_rgb: FastRed + FastBlue + DABgdx_from_rgb: Methyl Green + DABhax_from_rgb: Hematoxylin + AECbro_from_rgb: Blue matrix Anilline Blue + Red matrix Azocarmine + Orange matrix Orange-Gbpx_from_rgb: Methyl Blue + Ponceau Fuchsinahx_from_rgb: Alcian Blue + Hematoxylinhpx_from_rgb: Hematoxylin + PAS
This implementation borrows some ideas from DIPlib [2], e.g. the compensation using a small value to avoid log artifacts when calculating the Beer-Lambert law.
References
- 1
- 2
- 3
A. C. Ruifrok and D. A. Johnston, “Quantification of histochemical staining by color deconvolution,” Anal. Quant. Cytol. Histol., vol. 23, no. 4, pp. 291–299, Aug. 2001.
Examples
>>> import cupy as cp >>> from skimage import data >>> from cucim.skimage.color import separate_stains, hdx_from_rgb >>> ihc = cp.array(data.immunohistochemistry()) >>> ihc_hdx = separate_stains(ihc, hdx_from_rgb)
- cucim.skimage.color.xyz2lab(xyz, illuminant='D65', observer='2', *, channel_axis=-1)#
XYZ to CIE-LAB color space conversion.
- Parameters
- xyz(…, 3, …) array_like
The image in XYZ format. By default, the final dimension denotes channels.
- illuminant{“A”, “B”, “C”, “D50”, “D55”, “D65”, “D75”, “E”}, optional
The name of the illuminant (the function is NOT case sensitive).
- observer{“2”, “10”, “R”}, optional
One of: 2-degree observer, 10-degree observer, or ‘R’ observer as in R function grDevices::convertColor.
- channel_axisint, optional
This parameter indicates which axis of the array corresponds to channels.
- Returns
- out(…, 3, …) ndarray
The image in CIE-LAB format. Same dimensions as input.
- Raises
- ValueError
If xyz is not at least 2-D with shape (…, 3, …).
- ValueError
If either the illuminant or the observer angle is unsupported or unknown.
Notes
By default Observer=”2”, Illuminant=”D65”. CIE XYZ tristimulus values x_ref=95.047, y_ref=100., z_ref=108.883. See function get_xyz_coords for a list of supported illuminants.
References
Examples
>>> import cupy as cp >>> from skimage import data >>> from cucim.skimage.color import rgb2xyz, xyz2lab >>> img = cp.array(data.astronaut()) >>> img_xyz = rgb2xyz(img) >>> img_lab = xyz2lab(img_xyz)
- cucim.skimage.color.xyz2luv(xyz, illuminant='D65', observer='2', *, channel_axis=-1)#
XYZ to CIE-Luv color space conversion.
- Parameters
- xyz(…, 3, …) array_like
The image in XYZ format. By default, the final dimension denotes channels.
- illuminant{“A”, “B”, “C”, “D50”, “D55”, “D65”, “D75”, “E”}, optional
The name of the illuminant (the function is NOT case sensitive).
- observer{“2”, “10”, “R”}, optional
The aperture angle of the observer.
- channel_axisint, optional
This parameter indicates which axis of the array corresponds to channels.
- Returns
- out(…, 3, …) ndarray
The image in CIE-Luv format. Same dimensions as input.
- Raises
- ValueError
If xyz is not at least 2-D with shape (…, 3, …).
- ValueError
If either the illuminant or the observer angle are not supported or unknown.
Notes
By default XYZ conversion weights use observer=2A. Reference whitepoint for D65 Illuminant, with XYZ tristimulus values of
(95.047, 100., 108.883). See function ‘get_xyz_coords’ for a list of supported illuminants.References
Examples
>>> import cupy as cp >>> from skimage import data >>> from cucim.skimage.color import rgb2xyz, xyz2luv >>> img = cp.array(data.astronaut()) >>> img_xyz = rgb2xyz(img) >>> img_luv = xyz2luv(img_xyz)
- cucim.skimage.color.xyz2rgb(xyz, *, channel_axis=-1)#
XYZ to RGB color space conversion.
- Parameters
- xyz(…, 3, …) array_like
The image in XYZ format. By default, the final dimension denotes channels.
- channel_axisint, optional
This parameter indicates which axis of the array corresponds to channels.
- Returns
- out(…, 3, …) ndarray
The image in RGB format. Same dimensions as input.
- Raises
- ValueError
If xyz is not at least 2-D with shape (…, 3, …).
Notes
The CIE XYZ color space is derived from the CIE RGB color space. Note however that this function converts to sRGB.
References
Examples
>>> from skimage import data >>> from cucim.skimage.color import rgb2xyz, xyz2rgb >>> img = cp.array(data.astronaut()) >>> img_xyz = rgb2xyz(img) >>> img_rgb = xyz2rgb(img_xyz)
- cucim.skimage.color.ycbcr2rgb(ycbcr, *, channel_axis=-1)#
YCbCr to RGB color space conversion.
- Parameters
- ycbcr(…, 3, …) array_like
The image in YCbCr format. By default, the final dimension denotes channels.
- channel_axisint, optional
This parameter indicates which axis of the array corresponds to channels.
- Returns
- out(…, 3, …) ndarray
The image in RGB format. Same dimensions as input.
- Raises
- ValueError
If ycbcr is not at least 2-D with shape (…, 3, …).
Notes
Y is between 16 and 235. This is the color space commonly used by video codecs; it is sometimes incorrectly called “YUV”.
References
- cucim.skimage.color.ydbdr2rgb(ydbdr, *, channel_axis=-1)#
YDbDr to RGB color space conversion.
- Parameters
- ydbdr(…, 3, …) array_like
The image in YDbDr format. By default, the final dimension denotes channels.
- channel_axisint, optional
This parameter indicates which axis of the array corresponds to channels.
- Returns
- out(…, 3, …) ndarray
The image in RGB format. Same dimensions as input.
- Raises
- ValueError
If ydbdr is not at least 2-D with shape (…, 3, …).
Notes
This is the color space commonly used by video codecs, also called the reversible color transform in JPEG2000.
References
- cucim.skimage.color.yiq2rgb(yiq, *, channel_axis=-1)#
YIQ to RGB color space conversion.
- Parameters
- yiq(…, 3, …) array_like
The image in YIQ format. By default, the final dimension denotes channels.
- channel_axisint, optional
This parameter indicates which axis of the array corresponds to channels.
- Returns
- out(…, 3, …) ndarray
The image in RGB format. Same dimensions as input.
- Raises
- ValueError
If yiq is not at least 2-D with shape (…, 3, …).
- cucim.skimage.color.ypbpr2rgb(ypbpr, *, channel_axis=-1)#
YPbPr to RGB color space conversion.
- Parameters
- ypbpr(…, 3, …) array_like
The image in YPbPr format. By default, the final dimension denotes channels.
- channel_axisint, optional
This parameter indicates which axis of the array corresponds to channels.
- Returns
- out(…, 3) ndarray
The image in RGB format. Same dimensions as input.
- Raises
- ValueError
If ypbpr is not at least 2-D with shape (…, 3).
References
- cucim.skimage.color.yuv2rgb(yuv, *, channel_axis=-1)#
YUV to RGB color space conversion.
- Parameters
- yuv(…, 3, …) array_like
The image in YUV format. By default, the final dimension denotes channels.
- Returns
- out(…, 3, …) ndarray
The image in RGB format. Same dimensions as input.
- Raises
- ValueError
If yuv is not at least 2-D with shape (…, 3, …).
References
data#
- cucim.skimage.data.binary_blobs(length=512, blob_size_fraction=0.1, n_dim=2, volume_fraction=0.5, seed=None)#
Generate synthetic binary image with several rounded blob-like objects.
- Parameters
- lengthint, optional
Linear size of output image.
- blob_size_fractionfloat, optional
Typical linear size of blob, as a fraction of
length, should be smaller than 1.- n_dimint, optional
Number of dimensions of output image.
- volume_fractionfloat, default 0.5
Fraction of image pixels covered by the blobs (where the output is 1). Should be in [0, 1].
- seed{None, int, cupy.random.Generator}, optional
If seed is None the cupy.random.Generator singleton is used. If seed is an int, a new
Generatorinstance is used, seeded with seed. If seed is already aGeneratorinstance then that instance is used.
- Returns
- blobsndarray of bools
Output binary image
Notes
Warning: CuPy does not give identical randomly generated numbers as NumPy, so using a specific seed here will not give an identical pattern to the scikit-image implementation.
The behavior for a given random seed may also change across CuPy major versions. See: https://docs.cupy.dev/en/stable/reference/random.html
Examples
>>> from cucim.skimage import data >>> # tiny size (5, 5) >>> blobs = data.binary_blobs(length=5, blob_size_fraction=0.2, seed=1) >>> # larger size >>> blobs = data.binary_blobs(length=256, blob_size_fraction=0.1) >>> # Finer structures >>> blobs = data.binary_blobs(length=256, blob_size_fraction=0.05) >>> # Blobs cover a smaller volume fraction of the image >>> blobs = data.binary_blobs(length=256, volume_fraction=0.3)
exposure#
- cucim.skimage.exposure.adjust_gamma(image, gamma=1, gain=1)#
Performs Gamma Correction on the input image.
Also known as Power Law Transform. This function transforms the input image pixelwise according to the equation
O = I**gammaafter scaling each pixel to the range 0 to 1.- Parameters
- imagendarray
Input image.
- gammafloat, optional
Non negative real number. Default value is 1.
- gainfloat, optional
The constant multiplier. Default value is 1.
- Returns
- outndarray
Gamma corrected output image.
See also
Notes
For gamma greater than 1, the histogram will shift towards left and the output image will be darker than the input image.
For gamma less than 1, the histogram will shift towards right and the output image will be brighter than the input image.
References
Examples
>>> from skimage import data >>> from cucim.skimage import exposure, img_as_float >>> image = img_as_float(cp.array(data.moon())) >>> gamma_corrected = exposure.adjust_gamma(image, 2) >>> # Output is darker for gamma > 1 >>> image.mean() > gamma_corrected.mean() array(True)
- cucim.skimage.exposure.adjust_log(image, gain=1, inv=False)#
Performs Logarithmic correction on the input image.
This function transforms the input image pixelwise according to the equation
O = gain*log(1 + I)after scaling each pixel to the range 0 to 1.For inverse logarithmic correction, the equation is
O = gain*(2**I - 1).- Parameters
- imagendarray
Input image.
- gainfloat, optional
The constant multiplier. Default value is 1.
- invfloat, optional
If True, it performs inverse logarithmic correction, else correction will be logarithmic. Defaults to False.
- Returns
- outndarray
Logarithm corrected output image.
See also
References
- cucim.skimage.exposure.adjust_sigmoid(image, cutoff=0.5, gain=10, inv=False)#
Performs Sigmoid Correction on the input image.
Also known as Contrast Adjustment. This function transforms the input image pixelwise according to the equation
O = 1/(1 + exp*(gain*(cutoff - I)))after scaling each pixel to the range 0 to 1.- Parameters
- imagendarray
Input image.
- cutofffloat, optional
Cutoff of the sigmoid function that shifts the characteristic curve in horizontal direction. Default value is 0.5.
- gainfloat, optional
The constant multiplier in exponential’s power of sigmoid function. Default value is 10.
- invbool, optional
If True, returns the negative sigmoid correction. Defaults to False.
- Returns
- outndarray
Sigmoid corrected output image.
See also
References
- 1
Gustav J. Braun, “Image Lightness Rescaling Using Sigmoidal Contrast Enhancement Functions”, http://markfairchild.org/PDFs/PAP07.pdf
- cucim.skimage.exposure.cumulative_distribution(image, nbins=256)#
Return cumulative distribution function (cdf) for the given image.
- Parameters
- imagearray
Image array.
- nbinsint, optional
Number of bins for image histogram.
- Returns
- img_cdfarray
Values of cumulative distribution function.
- bin_centersarray
Centers of bins.
See also
References
Examples
>>> import cupy as cp >>> from skimage import data >>> from cucim.skimage import exposure, img_as_float >>> image = img_as_float(cp.array(data.camera())) >>> hi = exposure.histogram(image) >>> cdf = exposure.cumulative_distribution(image) >>> cp.alltrue(cdf[0] == cp.cumsum(hi[0])/float(image.size)) array(True)
- cucim.skimage.exposure.equalize_adapthist(image, kernel_size=None, clip_limit=0.01, nbins=256)#
Contrast Limited Adaptive Histogram Equalization (CLAHE).
An algorithm for local contrast enhancement, that uses histograms computed over different tile regions of the image. Local details can therefore be enhanced even in regions that are darker or lighter than most of the image.
- Parameters
- image(N1, …,NN[, C]) ndarray
Input image.
- kernel_sizeint or array_like, optional
Defines the shape of contextual regions used in the algorithm. If iterable is passed, it must have the same number of elements as
image.ndim(without color channel). If integer, it is broadcasted to each image dimension. By default,kernel_sizeis 1/8 ofimageheight by 1/8 of its width.- clip_limitfloat, optional
Clipping limit, normalized between 0 and 1 (higher values give more contrast).
- nbinsint, optional
Number of gray bins for histogram (“data range”).
- Returns
- out(N1, …,NN[, C]) ndarray
Equalized image with float64 dtype.
See also
Notes
- For color images, the following steps are performed:
The image is converted to HSV color space
The CLAHE algorithm is run on the V (Value) channel
The image is converted back to RGB space and returned
For RGBA images, the original alpha channel is removed.
Changed in version 0.17: The values returned by this function are slightly shifted upwards because of an internal change in rounding behavior.
References
- cucim.skimage.exposure.equalize_hist(image, nbins=256, mask=None)#
Return image after histogram equalization.
- Parameters
- imagearray
Image array.
- nbinsint, optional
Number of bins for image histogram. Note: this argument is ignored for integer images, for which each integer is its own bin.
- mask: ndarray of bools or 0s and 1s, optional
Array of same shape as image. Only points at which mask == True are used for the equalization, which is applied to the whole image.
- Returns
- outfloat array
Image array after histogram equalization.
Notes
This function is adapted from [1] with the author’s permission.
References
- cucim.skimage.exposure.histogram(image, nbins=256, source_range='image', normalize=False, *, channel_axis=None)#
Return histogram of image.
Unlike numpy.histogram, this function returns the centers of bins and does not rebin integer arrays. For integer arrays, each integer value has its own bin, which improves speed and intensity-resolution.
If channel_axis is not set, the histogram is computed on the flattened image. For color or multichannel images, set
channel_axisto use a common binning for all channels. Alternatively, one may apply the function separately on each channel to obtain a histogram for each color channel with separate binning.- Parameters
- imagearray
Input image.
- nbinsint, optional
Number of bins used to calculate histogram. This value is ignored for integer arrays.
- source_rangestring, optional
‘image’ (default) determines the range from the input image. ‘dtype’ determines the range from the expected range of the images of that data type.
- normalizebool, optional
If True, normalize the histogram by the sum of its values.
- channel_axisint or None, optional
If None, the image is assumed to be a grayscale (single channel) image. Otherwise, this parameter indicates which axis of the array corresponds to channels.
- Returns
- histarray
The values of the histogram. When
channel_axisis not None, hist will be a 2D array where the first axis corresponds to channels.- bin_centersarray
The values at the center of the bins.
See also
Examples
>>> import cupy as cp >>> from skimage import data >>> from cucim.skimage import exposure, img_as_float >>> image = img_as_float(cp.array(data.camera())) >>> cp.histogram(image, bins=2) (array([ 93585, 168559]), array([0. , 0.5, 1. ])) >>> exposure.histogram(image, nbins=2) (array([ 93585, 168559]), array([0.25, 0.75]))
- cucim.skimage.exposure.is_low_contrast(image, fraction_threshold=0.05, lower_percentile=1, upper_percentile=99, method='linear')#
Determine if an image is low contrast.
- Parameters
- imagearray-like
The image under test.
- fraction_thresholdfloat, optional
The low contrast fraction threshold. An image is considered low- contrast when its range of brightness spans less than this fraction of its data type’s full range. [1]
- lower_percentilefloat, optional
Disregard values below this percentile when computing image contrast.
- upper_percentilefloat, optional
Disregard values above this percentile when computing image contrast.
- methodstr, optional
The contrast determination method. Right now the only available option is “linear”.
- Returns
- outbool
True when the image is determined to be low contrast.
Notes
For boolean images, this function returns False only if all values are the same (the method, threshold, and percentile arguments are ignored).
References
Examples
>>> import cupy as cp >>> image = cp.linspace(0, 0.04, 100) >>> is_low_contrast(image) array(True) >>> image[-1] = 1 >>> is_low_contrast(image) array(True) >>> is_low_contrast(image, upper_percentile=100) array(False)
- cucim.skimage.exposure.match_histograms(image, reference, *, channel_axis=None, multichannel=False)#
Adjust an image so that its cumulative histogram matches that of another.
The adjustment is applied separately for each channel.
- Parameters
- imagendarray
Input image. Can be gray-scale or in color.
- referencendarray
Image to match histogram of. Must have the same number of channels as image.
- channel_axisint or None, optional
If None, the image is assumed to be a grayscale (single channel) image. Otherwise, this parameter indicates which axis of the array corresponds to channels.
- multichannelbool, optional
Apply the matching separately for each channel. This argument is deprecated: specify channel_axis instead.
- Returns
- matchedndarray
Transformed input image.
- Other Parameters
- multichannelDEPRECATED
Deprecated in favor of channel_axis.
Deprecated since version 22.02.00.
- Raises
- ValueError
Thrown when the number of channels in the input image and the reference differ.
References
- cucim.skimage.exposure.rescale_intensity(image, in_range='image', out_range='dtype')#
Return image after stretching or shrinking its intensity levels.
The desired intensity range of the input and output, in_range and out_range respectively, are used to stretch or shrink the intensity range of the input image. See examples below.
- Parameters
- imagearray
Image array.
- in_range, out_rangestr or 2-tuple, optional
Min and max intensity values of input and output image. The possible values for this parameter are enumerated below.
- ‘image’
Use image min/max as the intensity range.
- ‘dtype’
Use min/max of the image’s dtype as the intensity range.
- dtype-name
Use intensity range based on desired dtype. Must be valid key in DTYPE_RANGE.
- 2-tuple
Use range_values as explicit min/max intensities.
- Returns
- outarray
Image array after rescaling its intensity. This image is the same dtype as the input image.
See also
Notes
Changed in version 0.17: The dtype of the output array has changed to match the output dtype, or float if the output range is specified by a pair of floats.
Examples
By default, the min/max intensities of the input image are stretched to the limits allowed by the image’s dtype, since in_range defaults to ‘image’ and out_range defaults to ‘dtype’:
>>> image = cp.array([51, 102, 153], dtype=np.uint8) >>> rescale_intensity(image) array([ 0, 127, 255], dtype=uint8)
It’s easy to accidentally convert an image dtype from uint8 to float:
>>> 1.0 * image array([ 51., 102., 153.])
Use rescale_intensity to rescale to the proper range for float dtypes:
>>> image_float = 1.0 * image >>> rescale_intensity(image_float) array([0. , 0.5, 1. ])
To maintain the low contrast of the original, use the in_range parameter:
>>> rescale_intensity(image_float, in_range=(0, 255)) array([0.2, 0.4, 0.6])
If the min/max value of in_range is more/less than the min/max image intensity, then the intensity levels are clipped:
>>> rescale_intensity(image_float, in_range=(0, 102)) array([0.5, 1. , 1. ])
If you have an image with signed integers but want to rescale the image to just the positive range, use the out_range parameter. In that case, the output dtype will be float:
>>> image = cp.asarray([-10, 0, 10], dtype=np.int8) >>> rescale_intensity(image, out_range=(0, 127)) array([ 0. , 63.5, 127. ])
To get the desired range with a specific dtype, use
.astype():>>> rescale_intensity(image, out_range=(0, 127)).astype(np.int8) array([ 0, 63, 127], dtype=int8)
If the input image is constant, the output will be clipped directly to the output range: >>> image = cp.asarray([130, 130, 130], dtype=np.int32) >>> rescale_intensity(image, out_range=(0, 127)).astype(np.int32) array([127, 127, 127], dtype=int32)
feature#
- cucim.skimage.feature.canny(image, sigma=1.0, low_threshold=None, high_threshold=None, mask=None, use_quantiles=False, *, mode='constant', cval=0.0)#
Edge filter an image using the Canny algorithm.
- Parameters
- image2D array
Grayscale input image to detect edges on; can be of any dtype.
- sigmafloat, optional
Standard deviation of the Gaussian filter.
- low_thresholdfloat, optional
Lower bound for hysteresis thresholding (linking edges). If None, low_threshold is set to 10% of dtype’s max.
- high_thresholdfloat, optional
Upper bound for hysteresis thresholding (linking edges). If None, high_threshold is set to 20% of dtype’s max.
- maskarray, dtype=bool, optional
Mask to limit the application of Canny to a certain area.
- use_quantilesbool, optional
If
Truethen treat low_threshold and high_threshold as quantiles of the edge magnitude image, rather than absolute edge magnitude values. IfTruethen the thresholds must be in the range [0, 1].- modestr, {‘reflect’, ‘constant’, ‘nearest’, ‘mirror’, ‘wrap’}
The
modeparameter determines how the array borders are handled during Gaussian filtering, wherecvalis the value when mode is equal to ‘constant’.- cvalfloat, optional
Value to fill past edges of input if mode is ‘constant’.
- Returns
- output2D array (image)
The binary edge map.
See also
skimage.sobel
Notes
The steps of the algorithm are as follows:
Smooth the image using a Gaussian with
sigmawidth.Apply the horizontal and vertical Sobel operators to get the gradients within the image. The edge strength is the norm of the gradient.
Thin potential edges to 1-pixel wide curves. First, find the normal to the edge at each point. This is done by looking at the signs and the relative magnitude of the X-Sobel and Y-Sobel to sort the points into 4 categories: horizontal, vertical, diagonal and antidiagonal. Then look in the normal and reverse directions to see if the values in either of those directions are greater than the point in question. Use interpolation to get a mix of points instead of picking the one that’s the closest to the normal.
Perform a hysteresis thresholding: first label all points above the high threshold as edges. Then recursively label any point above the low threshold that is 8-connected to a labeled point as an edge.
References
- 1
Canny, J., A Computational Approach To Edge Detection, IEEE Trans. Pattern Analysis and Machine Intelligence, 8:679-714, 1986 DOI:10.1109/TPAMI.1986.4767851
- 2
William Green’s Canny tutorial https://en.wikipedia.org/wiki/Canny_edge_detector
Examples
>>> import cupy as cp >>> from cucim.skimage import feature >>> # Generate noisy image of a square >>> im = cp.zeros((256, 256)) >>> im[64:-64, 64:-64] = 1 >>> im += 0.2 * cp.random.rand(*im.shape) >>> # First trial with the Canny filter, with the default smoothing >>> edges1 = feature.canny(im) >>> # Increase the smoothing for better results >>> edges2 = feature.canny(im, sigma=3)
- cucim.skimage.feature.corner_foerstner(image, sigma=1)#
Compute Foerstner corner measure response image.
This corner detector uses information from the auto-correlation matrix A:
A = [(imx**2) (imx*imy)] = [Axx Axy] [(imx*imy) (imy**2)] [Axy Ayy]
Where imx and imy are first derivatives, averaged with a gaussian filter. The corner measure is then defined as:
w = det(A) / trace(A) (size of error ellipse) q = 4 * det(A) / trace(A)**2 (roundness of error ellipse)
- Parameters
- image(M, N) ndarray
Input image.
- sigmafloat, optional
Standard deviation used for the Gaussian kernel, which is used as weighting function for the auto-correlation matrix.
- Returns
- wndarray
Error ellipse sizes.
- qndarray
Roundness of error ellipse.
References
- 1
Förstner, W., & Gülch, E. (1987, June). A fast operator for detection and precise location of distinct points, corners and centres of circular features. In Proc. ISPRS intercommission conference on fast processing of photogrammetric data (pp. 281-305). https://cseweb.ucsd.edu/classes/sp02/cse252/foerstner/foerstner.pdf
- 2
Examples
>>> from cucim.skimage.feature import corner_foerstner, corner_peaks >>> square = cp.zeros([10, 10]) >>> square[2:8, 2:8] = 1 >>> square.astype(int) array([[0, 0, 0, 0, 0, 0, 0, 0, 0, 0], [0, 0, 0, 0, 0, 0, 0, 0, 0, 0], [0, 0, 1, 1, 1, 1, 1, 1, 0, 0], [0, 0, 1, 1, 1, 1, 1, 1, 0, 0], [0, 0, 1, 1, 1, 1, 1, 1, 0, 0], [0, 0, 1, 1, 1, 1, 1, 1, 0, 0], [0, 0, 1, 1, 1, 1, 1, 1, 0, 0], [0, 0, 1, 1, 1, 1, 1, 1, 0, 0], [0, 0, 0, 0, 0, 0, 0, 0, 0, 0], [0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]) >>> w, q = corner_foerstner(square) >>> accuracy_thresh = 0.5 >>> roundness_thresh = 0.3 >>> foerstner = (q > roundness_thresh) * (w > accuracy_thresh) * w >>> corner_peaks(foerstner, min_distance=1) array([[2, 2], [2, 7], [7, 2], [7, 7]])
- cucim.skimage.feature.corner_harris(image, method='k', k=0.05, eps=1e-06, sigma=1)#
Compute Harris corner measure response image.
This corner detector uses information from the auto-correlation matrix A:
A = [(imx**2) (imx*imy)] = [Axx Axy] [(imx*imy) (imy**2)] [Axy Ayy]
Where imx and imy are first derivatives, averaged with a gaussian filter. The corner measure is then defined as:
det(A) - k * trace(A)**2
or:
2 * det(A) / (trace(A) + eps)
- Parameters
- image(M, N) ndarray
Input image.
- method{‘k’, ‘eps’}, optional
Method to compute the response image from the auto-correlation matrix.
- kfloat, optional
Sensitivity factor to separate corners from edges, typically in range [0, 0.2]. Small values of k result in detection of sharp corners.
- epsfloat, optional
Normalisation factor (Noble’s corner measure).
- sigmafloat, optional
Standard deviation used for the Gaussian kernel, which is used as weighting function for the auto-correlation matrix.
- Returns
- responsendarray
Harris response image.
References
Examples
>>> from cucim.skimage.feature import corner_harris, corner_peaks >>> square = cp.zeros([10, 10]) >>> square[2:8, 2:8] = 1 >>> square.astype(int) array([[0, 0, 0, 0, 0, 0, 0, 0, 0, 0], [0, 0, 0, 0, 0, 0, 0, 0, 0, 0], [0, 0, 1, 1, 1, 1, 1, 1, 0, 0], [0, 0, 1, 1, 1, 1, 1, 1, 0, 0], [0, 0, 1, 1, 1, 1, 1, 1, 0, 0], [0, 0, 1, 1, 1, 1, 1, 1, 0, 0], [0, 0, 1, 1, 1, 1, 1, 1, 0, 0], [0, 0, 1, 1, 1, 1, 1, 1, 0, 0], [0, 0, 0, 0, 0, 0, 0, 0, 0, 0], [0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]) >>> corner_peaks(corner_harris(square), min_distance=1) array([[2, 2], [2, 7], [7, 2], [7, 7]])
- cucim.skimage.feature.corner_kitchen_rosenfeld(image, mode='constant', cval=0)#
Compute Kitchen and Rosenfeld corner measure response image.
The corner measure is calculated as follows:
(imxx * imy**2 + imyy * imx**2 - 2 * imxy * imx * imy) / (imx**2 + imy**2)
Where imx and imy are the first and imxx, imxy, imyy the second derivatives.
- Parameters
- image(M, N) ndarray
Input image.
- mode{‘constant’, ‘reflect’, ‘wrap’, ‘nearest’, ‘mirror’}, optional
How to handle values outside the image borders.
- cvalfloat, optional
Used in conjunction with mode ‘constant’, the value outside the image boundaries.
- Returns
- responsendarray
Kitchen and Rosenfeld response image.
References
- 1
Kitchen, L., & Rosenfeld, A. (1982). Gray-level corner detection. Pattern recognition letters, 1(2), 95-102. DOI:10.1016/0167-8655(82)90020-4
- cucim.skimage.feature.corner_peaks(image, min_distance=1, threshold_abs=None, threshold_rel=None, exclude_border=True, indices=True, num_peaks=inf, footprint=None, labels=None, *, num_peaks_per_label=inf, p_norm=inf)#
Find peaks in corner measure response image.
This differs from skimage.feature.peak_local_max in that it suppresses multiple connected peaks with the same accumulator value.
- Parameters
- image(M, N) ndarray
Input image.
- min_distanceint, optional
The minimal allowed distance separating peaks.
- **
See
skimage.feature.peak_local_max().- p_normfloat
Which Minkowski p-norm to use. Should be in the range [1, inf]. A finite large p may cause a ValueError if overflow can occur.
infcorresponds to the Chebyshev distance and 2 to the Euclidean distance.
- Returns
- outputndarray or ndarray of bools
If indices = True : (row, column, …) coordinates of peaks.
If indices = False : Boolean array shaped like image, with peaks represented by True values.
See also
Notes
Changed in version 0.18: The default value of threshold_rel has changed to None, which corresponds to letting skimage.feature.peak_local_max decide on the default. This is equivalent to threshold_rel=0.
The num_peaks limit is applied before suppression of connected peaks. To limit the number of peaks after suppression, set num_peaks=np.inf and post-process the output of this function.
Examples
>>> from cucim.skimage.feature import peak_local_max >>> response = cp.zeros((5, 5)) >>> response[2:4, 2:4] = 1 >>> response array([[0., 0., 0., 0., 0.], [0., 0., 0., 0., 0.], [0., 0., 1., 1., 0.], [0., 0., 1., 1., 0.], [0., 0., 0., 0., 0.]]) >>> peak_local_max(response) array([[2, 2], [2, 3], [3, 2], [3, 3]]) >>> corner_peaks(response) array([[2, 2]])
- cucim.skimage.feature.corner_shi_tomasi(image, sigma=1)#
Compute Shi-Tomasi (Kanade-Tomasi) corner measure response image.
This corner detector uses information from the auto-correlation matrix A:
A = [(imx**2) (imx*imy)] = [Axx Axy] [(imx*imy) (imy**2)] [Axy Ayy]
Where imx and imy are first derivatives, averaged with a gaussian filter. The corner measure is then defined as the smaller eigenvalue of A:
((Axx + Ayy) - sqrt((Axx - Ayy)**2 + 4 * Axy**2)) / 2
- Parameters
- image(M, N) ndarray
Input image.
- sigmafloat, optional
Standard deviation used for the Gaussian kernel, which is used as weighting function for the auto-correlation matrix.
- Returns
- responsendarray
Shi-Tomasi response image.
References
Examples
>>> from cucim.skimage.feature import corner_shi_tomasi, corner_peaks >>> square = cp.zeros([10, 10]) >>> square[2:8, 2:8] = 1 >>> square.astype(int) array([[0, 0, 0, 0, 0, 0, 0, 0, 0, 0], [0, 0, 0, 0, 0, 0, 0, 0, 0, 0], [0, 0, 1, 1, 1, 1, 1, 1, 0, 0], [0, 0, 1, 1, 1, 1, 1, 1, 0, 0], [0, 0, 1, 1, 1, 1, 1, 1, 0, 0], [0, 0, 1, 1, 1, 1, 1, 1, 0, 0], [0, 0, 1, 1, 1, 1, 1, 1, 0, 0], [0, 0, 1, 1, 1, 1, 1, 1, 0, 0], [0, 0, 0, 0, 0, 0, 0, 0, 0, 0], [0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]) >>> corner_peaks(corner_shi_tomasi(square), ... min_distance=1) array([[2, 2], [2, 7], [7, 2], [7, 7]])
- cucim.skimage.feature.daisy(image, step=4, radius=15, rings=3, histograms=8, orientations=8, normalization='l1', sigmas=None, ring_radii=None, visualize=False)#
Extract DAISY feature descriptors densely for the given image.
DAISY is a feature descriptor similar to SIFT formulated in a way that allows for fast dense extraction. Typically, this is practical for bag-of-features image representations.
The implementation follows Tola et al. [1] but deviate on the following points:
Histogram bin contribution are smoothed with a circular Gaussian window over the tonal range (the angular range).
The sigma values of the spatial Gaussian smoothing in this code do not match the sigma values in the original code by Tola et al. [2]. In their code, spatial smoothing is applied to both the input image and the center histogram. However, this smoothing is not documented in [1] and, therefore, it is omitted.
- Parameters
- image(M, N) array
Input image (grayscale).
- stepint, optional
Distance between descriptor sampling points.
- radiusint, optional
Radius (in pixels) of the outermost ring.
- ringsint, optional
Number of rings.
- histogramsint, optional
Number of histograms sampled per ring.
- orientationsint, optional
Number of orientations (bins) per histogram.
- normalization[ ‘l1’ | ‘l2’ | ‘daisy’ | ‘off’ ], optional
How to normalize the descriptors
‘l1’: L1-normalization of each descriptor.
‘l2’: L2-normalization of each descriptor.
‘daisy’: L2-normalization of individual histograms.
‘off’: Disable normalization.
- sigmas1D array of float, optional
Standard deviation of spatial Gaussian smoothing for the center histogram and for each ring of histograms. The array of sigmas should be sorted from the center and out. I.e. the first sigma value defines the spatial smoothing of the center histogram and the last sigma value defines the spatial smoothing of the outermost ring. Specifying sigmas overrides the following parameter.
rings = len(sigmas) - 1- ring_radii1D array of int, optional
Radius (in pixels) for each ring. Specifying ring_radii overrides the following two parameters.
rings = len(ring_radii)radius = ring_radii[-1]If both sigmas and ring_radii are given, they must satisfy the following predicate since no radius is needed for the center histogram.
len(ring_radii) == len(sigmas) + 1- visualizebool, optional
Generate a visualization of the DAISY descriptors
- Returns
- descsarray
Grid of DAISY descriptors for the given image as an array dimensionality (P, Q, R) where
P = ceil((M - radius*2) / step)Q = ceil((N - radius*2) / step)R = (rings * histograms + 1) * orientations- descs_img(M, N, 3) array (only if visualize==True)
Visualization of the DAISY descriptors.
References
- cucim.skimage.feature.hessian_matrix(image, sigma=1, mode='constant', cval=0, order='rc')#
Compute the Hessian matrix.
In 2D, the Hessian matrix is defined as:
H = [Hrr Hrc] [Hrc Hcc]
which is computed by convolving the image with the second derivatives of the Gaussian kernel in the respective r- and c-directions.
The implementation here also supports n-dimensional data.
- Parameters
- imagendarray
Input image.
- sigmafloat
Standard deviation used for the Gaussian kernel, which is used as weighting function for the auto-correlation matrix.
- mode{‘constant’, ‘reflect’, ‘wrap’, ‘nearest’, ‘mirror’}, optional
How to handle values outside the image borders.
- cvalfloat, optional
Used in conjunction with mode ‘constant’, the value outside the image boundaries.
- order{‘rc’, ‘xy’}, optional
This parameter allows for the use of reverse or forward order of the image axes in gradient computation. ‘rc’ indicates the use of the first axis initially (Hrr, Hrc, Hcc), whilst ‘xy’ indicates the usage of the last axis initially (Hxx, Hxy, Hyy)
- Returns
- H_elemslist of ndarray
Upper-diagonal elements of the hessian matrix for each pixel in the input image. In 2D, this will be a three element list containing [Hrr, Hrc, Hcc]. In nD, the list will contain
(n**2 + n) / 2arrays.
Examples
>>> import cupy as cp >>> from cucim.skimage.feature import hessian_matrix >>> square = cp.zeros((5, 5)) >>> square[2, 2] = 4 >>> Hrr, Hrc, Hcc = hessian_matrix(square, sigma=0.1, order='rc') >>> Hrc array([[ 0., 0., 0., 0., 0.], [ 0., 1., 0., -1., 0.], [ 0., 0., 0., 0., 0.], [ 0., -1., 0., 1., 0.], [ 0., 0., 0., 0., 0.]])
- cucim.skimage.feature.hessian_matrix_det(image, sigma=1, approximate=True)#
Compute the approximate Hessian Determinant over an image.
The 2D approximate method uses box filters over integral images to compute the approximate Hessian Determinant.
- Parameters
- imagendarray
The image over which to compute the Hessian Determinant.
- sigmafloat, optional
Standard deviation of the Gaussian kernel used for the Hessian matrix.
- approximatebool, optional
If
Trueand the image is 2D, use a much faster approximate computation. This argument has no effect on 3D and higher images.
- Returns
- outarray
The array of the Determinant of Hessians.
Notes
For 2D images when
approximate=True, the running time of this method only depends on size of the image. It is independent of sigma as one would expect. The downside is that the result for sigma less than 3 is not accurate, i.e., not similar to the result obtained if someone computed the Hessian and took its determinant.References
- 1
Herbert Bay, Andreas Ess, Tinne Tuytelaars, Luc Van Gool, “SURF: Speeded Up Robust Features” ftp://ftp.vision.ee.ethz.ch/publications/articles/eth_biwi_00517.pdf
- cucim.skimage.feature.hessian_matrix_eigvals(H_elems)#
Compute eigenvalues of Hessian matrix.
- Parameters
- H_elemslist of ndarray
The upper-diagonal elements of the Hessian matrix, as returned by hessian_matrix.
- Returns
- eigsndarray
The eigenvalues of the Hessian matrix, in decreasing order. The eigenvalues are the leading dimension. That is,
eigs[i, j, k]contains the ith-largest eigenvalue at position (j, k).
Examples
>>> import cupy as cp >>> from cucim.skimage.feature import (hessian_matrix, ... hessian_matrix_eigvals) >>> square = cp.zeros((5, 5)) >>> square[2, 2] = 4 >>> H_elems = hessian_matrix(square, sigma=0.1, order='rc') >>> hessian_matrix_eigvals(H_elems)[0] array([[ 0., 0., 2., 0., 0.], [ 0., 1., 0., 1., 0.], [ 2., 0., -2., 0., 2.], [ 0., 1., 0., 1., 0.], [ 0., 0., 2., 0., 0.]])
- cucim.skimage.feature.masked_register_translation(src_image, target_image, src_mask, target_mask=None, overlap_ratio=0.3)#
Deprecated function. Use
cucim.skimage.registration.phase_cross_correlationinstead.
- cucim.skimage.feature.match_template(image, template, pad_input=False, mode='constant', constant_values=0)#
Match a template to a 2-D or 3-D image using normalized correlation.
The output is an array with values between -1.0 and 1.0. The value at a given position corresponds to the correlation coefficient between the image and the template.
For pad_input=True matches correspond to the center and otherwise to the top-left corner of the template. To find the best match you must search for peaks in the response (output) image.
- Parameters
- image(M, N[, D]) array
2-D or 3-D input image.
- template(m, n[, d]) array
Template to locate. It must be (m <= M, n <= N[, d <= D]).
- pad_inputbool
If True, pad image so that output is the same size as the image, and output values correspond to the template center. Otherwise, the output is an array with shape (M - m + 1, N - n + 1) for an (M, N) image and an (m, n) template, and matches correspond to origin (top-left corner) of the template.
- modesee numpy.pad, optional
Padding mode.
- constant_valuessee numpy.pad, optional
Constant values used in conjunction with
mode='constant'.
- Returns
- outputarray
Response image with correlation coefficients.
Notes
Details on the cross-correlation are presented in [1]. This implementation uses FFT convolutions of the image and the template. Reference [2] presents similar derivations but the approximation presented in this reference is not used in our implementation.
This CuPy implementation does not force the image to float64 internally, but will use float32 for single-precision inputs.
References
- 1
J. P. Lewis, “Fast Normalized Cross-Correlation”, Industrial Light and Magic.
- 2
Briechle and Hanebeck, “Template Matching using Fast Normalized Cross Correlation”, Proceedings of the SPIE (2001). DOI:10.1117/12.421129
Examples
>>> import cupy as cp >>> template = cp.zeros((3, 3)) >>> template[1, 1] = 1 >>> template array([[0., 0., 0.], [0., 1., 0.], [0., 0., 0.]]) >>> image = cp.zeros((6, 6)) >>> image[1, 1] = 1 >>> image[4, 4] = -1 >>> image array([[ 0., 0., 0., 0., 0., 0.], [ 0., 1., 0., 0., 0., 0.], [ 0., 0., 0., 0., 0., 0.], [ 0., 0., 0., 0., 0., 0.], [ 0., 0., 0., 0., -1., 0.], [ 0., 0., 0., 0., 0., 0.]]) >>> result = match_template(image, template) >>> cp.round(result, 3) array([[ 1. , -0.125, 0. , 0. ], [-0.125, -0.125, 0. , 0. ], [ 0. , 0. , 0.125, 0.125], [ 0. , 0. , 0.125, -1. ]]) >>> result = match_template(image, template, pad_input=True) >>> cp.round(result, 3) array([[-0.125, -0.125, -0.125, 0. , 0. , 0. ], [-0.125, 1. , -0.125, 0. , 0. , 0. ], [-0.125, -0.125, -0.125, 0. , 0. , 0. ], [ 0. , 0. , 0. , 0.125, 0.125, 0.125], [ 0. , 0. , 0. , 0.125, -1. , 0.125], [ 0. , 0. , 0. , 0.125, 0.125, 0.125]])
- cucim.skimage.feature.multiscale_basic_features(image, multichannel=False, intensity=True, edges=True, texture=True, sigma_min=0.5, sigma_max=16, num_sigma=None, num_workers=None, *, channel_axis=None)#
Local features for a single- or multi-channel nd image.
Intensity, gradient intensity and local structure are computed at different scales thanks to Gaussian blurring.
- Parameters
- imagendarray
Input image, which can be grayscale or multichannel.
- multichannelbool, default False
True if the last dimension corresponds to color channels. This argument is deprecated: specify channel_axis instead.
- intensitybool, default True
If True, pixel intensities averaged over the different scales are added to the feature set.
- edgesbool, default True
If True, intensities of local gradients averaged over the different scales are added to the feature set.
- texturebool, default True
If True, eigenvalues of the Hessian matrix after Gaussian blurring at different scales are added to the feature set.
- sigma_minfloat, optional
Smallest value of the Gaussian kernel used to average local neighbourhoods before extracting features.
- sigma_maxfloat, optional
Largest value of the Gaussian kernel used to average local neighbourhoods before extracting features.
- num_sigmaint, optional
Number of values of the Gaussian kernel between sigma_min and sigma_max. If None, sigma_min multiplied by powers of 2 are used.
- num_workersint or None, optional
The number of parallel threads to use. If set to
None, the full set of available cores are used.- channel_axisint or None, optional
If None, the image is assumed to be a grayscale (single channel) image. Otherwise, this parameter indicates which axis of the array corresponds to channels.
- Returns
- featuresnp.ndarray
Array of shape
image.shape + (n_features,). When channel_axis is not None, all channels are concatenated along the features dimension. (i.e.n_features == n_features_singlechannel * n_channels)
- Other Parameters
- multichannelDEPRECATED
Deprecated in favor of channel_axis.
Deprecated since version 22.02.00.
- cucim.skimage.feature.peak_local_max(image, min_distance=1, threshold_abs=None, threshold_rel=None, exclude_border=True, indices=True, num_peaks=inf, footprint=None, labels=None, num_peaks_per_label=inf, p_norm=inf)#
Find peaks in an image as coordinate list or boolean mask.
Peaks are the local maxima in a region of 2 * min_distance + 1 (i.e. peaks are separated by at least min_distance).
If both threshold_abs and threshold_rel are provided, the maximum of the two is chosen as the minimum intensity threshold of peaks.
Changed in version 0.18: Prior to version 0.18, peaks of the same height within a radius of min_distance were all returned, but this could cause unexpected behaviour. From 0.18 onwards, an arbitrary peak within the region is returned. See issue gh-2592.
- Parameters
- imagendarray
Input image.
- min_distanceint, optional
The minimal allowed distance separating peaks. To find the maximum number of peaks, use min_distance=1.
- threshold_absfloat or None, optional
Minimum intensity of peaks. By default, the absolute threshold is the minimum intensity of the image.
- threshold_relfloat or None, optional
Minimum intensity of peaks, calculated as
max(image) * threshold_rel.- exclude_borderint, tuple of ints, or bool, optional
If positive integer, exclude_border excludes peaks from within exclude_border-pixels of the border of the image. If tuple of non-negative ints, the length of the tuple must match the input array’s dimensionality. Each element of the tuple will exclude peaks from within exclude_border-pixels of the border of the image along that dimension. If True, takes the min_distance parameter as value. If zero or False, peaks are identified regardless of their distance from the border.
- indicesbool, optional
If True, the output will be an array representing peak coordinates. The coordinates are sorted according to peaks values (Larger first). If False, the output will be a boolean array shaped as image.shape with peaks present at True elements.
indicesis deprecated and will be removed in version 0.20. Default behavior will be to always return peak coordinates. You can obtain a mask as shown in the example below.- num_peaksint, optional
Maximum number of peaks. When the number of peaks exceeds num_peaks, return num_peaks peaks based on highest peak intensity.
- footprintndarray of bools, optional
If provided, footprint == 1 represents the local region within which to search for peaks at every point in image.
- labelsndarray of ints, optional
If provided, each unique region labels == value represents a unique region to search for peaks. Zero is reserved for background.
- num_peaks_per_labelint, optional
Maximum number of peaks for each label.
- p_normfloat
Which Minkowski p-norm to use. Should be in the range [1, inf]. A finite large p may cause a ValueError if overflow can occur.
infcorresponds to the Chebyshev distance and 2 to the Euclidean distance.
- Returns
- outputndarray or ndarray of bools
If indices = True : (row, column, …) coordinates of peaks.
If indices = False : Boolean array shaped like image, with peaks represented by True values.
See also
Notes
The peak local maximum function returns the coordinates of local peaks (maxima) in an image. Internally, a maximum filter is used for finding local maxima. This operation dilates the original image. After comparison of the dilated and original image, this function returns the coordinates or a mask of the peaks where the dilated image equals the original image.
Examples
>>> import cupy as cp >>> img1 = cp.zeros((7, 7)) >>> img1[3, 4] = 1 >>> img1[3, 2] = 1.5 >>> img1 array([[0. , 0. , 0. , 0. , 0. , 0. , 0. ], [0. , 0. , 0. , 0. , 0. , 0. , 0. ], [0. , 0. , 0. , 0. , 0. , 0. , 0. ], [0. , 0. , 1.5, 0. , 1. , 0. , 0. ], [0. , 0. , 0. , 0. , 0. , 0. , 0. ], [0. , 0. , 0. , 0. , 0. , 0. , 0. ], [0. , 0. , 0. , 0. , 0. , 0. , 0. ]])
>>> peak_local_max(img1, min_distance=1) array([[3, 2], [3, 4]])
>>> peak_local_max(img1, min_distance=2) array([[3, 2]])
>>> img2 = cp.zeros((20, 20, 20)) >>> img2[10, 10, 10] = 1 >>> img2[15, 15, 15] = 1 >>> peak_idx = peak_local_max(img2, exclude_border=0) >>> peak_idx array([[10, 10, 10], [15, 15, 15]])
>>> peak_mask = cp.zeros_like(img2, dtype=bool) >>> peak_mask[tuple(peak_idx.T)] = True >>> np.argwhere(peak_mask) array([[10, 10, 10], [15, 15, 15]])
- cucim.skimage.feature.register_translation(src_image, target_image, upsample_factor=1, space='real', return_error=True)#
Deprecated function. Use
cucim.skimage.registration.phase_cross_correlationinstead.
- cucim.skimage.feature.shape_index(image, sigma=1, mode='constant', cval=0)#
Compute the shape index.
The shape index, as defined by Koenderink & van Doorn [1], is a single valued measure of local curvature, assuming the image as a 3D plane with intensities representing heights.
It is derived from the eigenvalues of the Hessian, and its value ranges from -1 to 1 (and is undefined (=NaN) in flat regions), with following ranges representing following shapes:
Ranges of the shape index and corresponding shapes.# Interval (s in …)
Shape
[ -1, -7/8)
Spherical cup
[-7/8, -5/8)
Through
[-5/8, -3/8)
Rut
[-3/8, -1/8)
Saddle rut
[-1/8, +1/8)
Saddle
[+1/8, +3/8)
Saddle ridge
[+3/8, +5/8)
Ridge
[+5/8, +7/8)
Dome
[+7/8, +1]
Spherical cap
- Parameters
- image(M, N) ndarray
Input image.
- sigmafloat, optional
Standard deviation used for the Gaussian kernel, which is used for smoothing the input data before Hessian eigen value calculation.
- mode{‘constant’, ‘reflect’, ‘wrap’, ‘nearest’, ‘mirror’}, optional
How to handle values outside the image borders
- cvalfloat, optional
Used in conjunction with mode ‘constant’, the value outside the image boundaries.
- Returns
- sndarray
Shape index
References
- 1
Koenderink, J. J. & van Doorn, A. J., “Surface shape and curvature scales”, Image and Vision Computing, 1992, 10, 557-564. DOI:10.1016/0262-8856(92)90076-F
Examples
>>> from cucim.skimage.feature import shape_index >>> square = cp.zeros((5, 5)) >>> square[2, 2] = 4 >>> s = shape_index(square, sigma=0.1) >>> s array([[ nan, nan, -0.5, nan, nan], [ nan, -0. , nan, -0. , nan], [-0.5, nan, -1. , nan, -0.5], [ nan, -0. , nan, -0. , nan], [ nan, nan, -0.5, nan, nan]])
- cucim.skimage.feature.structure_tensor(image, sigma=1, mode='constant', cval=0, order=None)#
Compute structure tensor using sum of squared differences.
The (2-dimensional) structure tensor A is defined as:
A = [Arr Arc] [Arc Acc]
which is approximated by the weighted sum of squared differences in a local window around each pixel in the image. This formula can be extended to a larger number of dimensions (see [1]).
- Parameters
- imagendarray
Input image.
- sigmafloat or array-like of float, optional
Standard deviation used for the Gaussian kernel, which is used as a weighting function for the local summation of squared differences. If sigma is an iterable, its length must be equal to image.ndim and each element is used for the Gaussian kernel applied along its respective axis.
- mode{‘constant’, ‘reflect’, ‘wrap’, ‘nearest’, ‘mirror’}, optional
How to handle values outside the image borders.
- cvalfloat, optional
Used in conjunction with mode ‘constant’, the value outside the image boundaries.
- order{‘rc’, ‘xy’}, optional
NOTE: Only applies in 2D. Higher dimensions must always use ‘rc’ order. This parameter allows for the use of reverse or forward order of the image axes in gradient computation. ‘rc’ indicates the use of the first axis initially (Arr, Arc, Acc), whilst ‘xy’ indicates the usage of the last axis initially (Axx, Axy, Ayy).
- Returns
- A_elemslist of ndarray
Upper-diagonal elements of the structure tensor for each pixel in the input image.
See also
References
Examples
>>> import cupy as cp >>> from cucim.skimage.feature import structure_tensor >>> square = cp.zeros((5, 5)) >>> square[2, 2] = 1 >>> Arr, Arc, Acc = structure_tensor(square, sigma=0.1, order="rc") >>> Acc array([[0., 0., 0., 0., 0.], [0., 1., 0., 1., 0.], [0., 4., 0., 4., 0.], [0., 1., 0., 1., 0.], [0., 0., 0., 0., 0.]])
- cucim.skimage.feature.structure_tensor_eigenvalues(A_elems)#
Compute eigenvalues of structure tensor.
- Parameters
- A_elemslist of ndarray
The upper-diagonal elements of the structure tensor, as returned by structure_tensor.
- Returns
- ndarray
The eigenvalues of the structure tensor, in decreasing order. The eigenvalues are the leading dimension. That is, the coordinate [i, j, k] corresponds to the ith-largest eigenvalue at position (j, k).
See also
Examples
>>> import cupy as cp >>> from cucim.skimage.feature import structure_tensor >>> from cucim.skimage.feature import structure_tensor_eigenvalues >>> square = cp.zeros((5, 5)) >>> square[2, 2] = 1 >>> A_elems = structure_tensor(square, sigma=0.1, order='rc') >>> structure_tensor_eigenvalues(A_elems)[0] array([[0., 0., 0., 0., 0.], [0., 2., 4., 2., 0.], [0., 4., 0., 4., 0.], [0., 2., 4., 2., 0.], [0., 0., 0., 0., 0.]])
filters#
- class cucim.skimage.filters.LPIFilter2D(impulse_response, **filter_params)#
Linear Position-Invariant Filter (2-dimensional)
Methods
__call__(data)Apply the filter to the given data.
- cucim.skimage.filters.apply_hysteresis_threshold(image, low, high)#
Apply hysteresis thresholding to
image.This algorithm finds regions where
imageis greater thanhighORimageis greater thanlowand that region is connected to a region greater thanhigh.- Parameters
- imagearray, shape (M,[ N, …, P])
Grayscale input image.
- lowfloat, or array of same shape as
image Lower threshold.
- highfloat, or array of same shape as
image Higher threshold.
- Returns
- thresholdedarray of bool, same shape as
image Array in which
Trueindicates the locations whereimagewas above the hysteresis threshold.
- thresholdedarray of bool, same shape as
References
- 1
J. Canny. A computational approach to edge detection. IEEE Transactions on Pattern Analysis and Machine Intelligence. 1986; vol. 8, pp.679-698. DOI:10.1109/TPAMI.1986.4767851
Examples
>>> import cupy as cp >>> from cucim.skimage.filters import apply_hysteresis_threshold >>> image = cp.asarray([1, 2, 3, 2, 1, 2, 1, 3, 2]) >>> apply_hysteresis_threshold(image, 1.5, 2.5).astype(int) array([0, 1, 1, 1, 0, 0, 0, 1, 1])
- cucim.skimage.filters.butterworth(image, cutoff_frequency_ratio=0.005, high_pass=True, order=2.0, channel_axis=None)#
Apply a Butterworth filter to enhance high or low frequency features.
This filter is defined in the Fourier domain.
- Parameters
- image(M[, N[, …, P]][, C]) ndarray
Input image.
- cutoff_frequency_ratiofloat, optional
Determines the position of the cut-off relative to the shape of the FFT.
- high_passbool, optional
Whether to perform a high pass filter. If False, a low pass filter is performed.
- orderfloat, optional
Order of the filter which affects the slope near the cut-off. Higher order means steeper slope in frequency space.
- channel_axisint, optional
If there is a channel dimension, provide the index here. If None (default) then all axes are assumed to be spatial dimensions.
- Returns
- resultndarray
The Butterworth-filtered image.
Notes
A band-pass filter can be achieved by combining a high pass and low pass filter.
The literature contains multiple conventions for the functional form of the Butterworth filter. Here it is implemented as the n-dimensional form of
\[\frac{1}{1 - \left(\frac{f}{c*f_{max}}\right)^{2*n}}\]with \(f\) the absolute value of the spatial frequency, \(c\) the
cutoff_frequency_ratioand \(n\) theordermodeled after [2]References
- 1
Butterworth, Stephen. “On the theory of filter amplifiers.” Wireless Engineer 7.6 (1930): 536-541.
- 2
Russ, John C., et al. “The image processing handbook.” Computers in Physics 8.2 (1994): 177-178.
Examples
Apply a high pass and low pass Butterworth filter to a grayscale and color image respectively:
>>> import cupy as cp >>> from skimage.data import camera, astronaut >>> from cucim.skimage.filters import butterworth >>> cam = cp.asarray(camera()) >>> astro = cp.asarray(astronaut()) >>> high_pass = butterworth(cam, 0.07, True, 8) >>> low_pass = butterworth(astro, 0.01, False, 4, channel_axis=-1)
- cucim.skimage.filters.correlate_sparse(image, kernel, mode='reflect')#
Compute valid cross-correlation of padded_array and kernel.
This function is fast when kernel is large with many zeros.
See
scipy.ndimage.correlatefor a description of cross-correlation.- Parameters
- imagendarray, dtype float, shape (M, N,[ …,] P)
The input array. If mode is ‘valid’, this array should already be padded, as a margin of the same shape as kernel will be stripped off.
- kernelndarray, dtype float shape (Q, R,[ …,] S)
The kernel to be correlated. Must have the same number of dimensions as padded_array. For high performance, it should be sparse (few nonzero entries).
- modestring, optional
See scipy.ndimage.correlate for valid modes. Additionally, mode ‘valid’ is accepted, in which case no padding is applied and the result is the result for the smaller image for which the kernel is entirely inside the original data.
- Returns
- resultarray of float, shape (M, N,[ …,] P)
The result of cross-correlating image with kernel. If mode ‘valid’ is used, the resulting shape is (M-Q+1, N-R+1,[ …,] P-S+1).
- cucim.skimage.filters.difference_of_gaussians(image, low_sigma, high_sigma=None, *, mode='nearest', cval=0, channel_axis=None, truncate=4.0, multichannel=False)#
Find features between
low_sigmaandhigh_sigmain size.This function uses the Difference of Gaussians method for applying band-pass filters to multi-dimensional arrays. The input array is blurred with two Gaussian kernels of differing sigmas to produce two intermediate, filtered images. The more-blurred image is then subtracted from the less-blurred image. The final output image will therefore have had high-frequency components attenuated by the smaller-sigma Gaussian, and low frequency components will have been removed due to their presence in the more-blurred intermediate.
- Parameters
- imagendarray
Input array to filter.
- low_sigmascalar or sequence of scalars
Standard deviation(s) for the Gaussian kernel with the smaller sigmas across all axes. The standard deviations are given for each axis as a sequence, or as a single number, in which case the single number is used as the standard deviation value for all axes.
- high_sigmascalar or sequence of scalars, optional (default is None)
Standard deviation(s) for the Gaussian kernel with the larger sigmas across all axes. The standard deviations are given for each axis as a sequence, or as a single number, in which case the single number is used as the standard deviation value for all axes. If None is given (default), sigmas for all axes are calculated as 1.6 * low_sigma.
- mode{‘reflect’, ‘constant’, ‘nearest’, ‘mirror’, ‘wrap’}, optional
The
modeparameter determines how the array borders are handled, wherecvalis the value when mode is equal to ‘constant’. Default is ‘nearest’.- cvalscalar, optional
Value to fill past edges of input if
modeis ‘constant’. Default is 0.0- channel_axisint or None, optional
If None, the image is assumed to be a grayscale (single channel) image. Otherwise, this parameter indicates which axis of the array corresponds to channels.
- truncatefloat, optional (default is 4.0)
Truncate the filter at this many standard deviations.
- multichannelbool, optional (default: False)
Whether the last axis of the image is to be interpreted as multiple channels. If True, each channel is filtered separately (channels are not mixed together). This argument is deprecated: specify channel_axis instead.
- Returns
- filtered_imagendarray
the filtered array.
- Other Parameters
- multichannelDEPRECATED
Deprecated in favor of channel_axis.
Deprecated since version 22.02.00.
See also
skimage.feature.blog_dog()
Notes
This function will subtract an array filtered with a Gaussian kernel with sigmas given by
high_sigmafrom an array filtered with a Gaussian kernel with sigmas provided bylow_sigma. The values forhigh_sigmamust always be greater than or equal to the corresponding values inlow_sigma, or aValueErrorwill be raised.When
high_sigmais none, the values forhigh_sigmawill be calculated as 1.6x the corresponding values inlow_sigma. This ratio was originally proposed by Marr and Hildreth (1980) [1] and is commonly used when approximating the inverted Laplacian of Gaussian, which is used in edge and blob detection.Input image is converted according to the conventions of
img_as_float.Except for sigma values, all parameters are used for both filters.
References
- 1
Marr, D. and Hildreth, E. Theory of Edge Detection. Proc. R. Soc. Lond. Series B 207, 187-217 (1980). https://doi.org/10.1098/rspb.1980.0020
Examples
Apply a simple Difference of Gaussians filter to a color image:
>>> from skimage.data import astronaut >>> from cucim.skimage.filters import difference_of_gaussians >>> astro = cp.asarray(astronaut()) >>> filtered_image = difference_of_gaussians(astro, 2, 10, ... multichannel=True)
Apply a Laplacian of Gaussian filter as approximated by the Difference of Gaussians filter:
>>> filtered_image = difference_of_gaussians(astro, 2, ... multichannel=True)
Apply a Difference of Gaussians filter to a grayscale image using different sigma values for each axis:
>>> from skimage.data import camera >>> cam = cp.array(camera()) >>> filtered_image = difference_of_gaussians(cam, (2,5), (3,20))
- cucim.skimage.filters.farid(image, *, mask=None)#
Find the edge magnitude using the Farid transform.
- Parameters
- image2-D array
Image to process.
- mask2-D array, optional
An optional mask to limit the application to a certain area. Note that pixels surrounding masked regions are also masked to prevent masked regions from affecting the result.
- Returns
- output2-D array
The Farid edge map.
See also
Notes
Take the square root of the sum of the squares of the horizontal and vertical derivatives to get a magnitude that is somewhat insensitive to direction. Similar to the Scharr operator, this operator is designed with a rotation invariance constraint.
References
- 1
Farid, H. and Simoncelli, E. P., “Differentiation of discrete multidimensional signals”, IEEE Transactions on Image Processing 13(4): 496-508, 2004. DOI:10.1109/TIP.2004.823819
- 2
Wikipedia, “Farid and Simoncelli Derivatives.” Available at: <https://en.wikipedia.org/wiki/Image_derivatives#Farid_and_Simoncelli_Derivatives>
Examples
>>> import cupy as cp >>> from skimage import data >>> camera = cp.array(data.camera()) >>> from cucim.skimage import filters >>> edges = filters.farid(camera)
- cucim.skimage.filters.farid_h(image, *, mask=None)#
Find the horizontal edges of an image using the Farid transform.
- Parameters
- image2-D array
Image to process.
- mask2-D array, optional
An optional mask to limit the application to a certain area. Note that pixels surrounding masked regions are also masked to prevent masked regions from affecting the result.
- Returns
- output2-D array
The Farid edge map.
Notes
The kernel was constructed using the 5-tap weights from [1].
References
- 1
Farid, H. and Simoncelli, E. P., “Differentiation of discrete multidimensional signals”, IEEE Transactions on Image Processing 13(4): 496-508, 2004. DOI:10.1109/TIP.2004.823819
- 2
Farid, H. and Simoncelli, E. P. “Optimally rotation-equivariant directional derivative kernels”, In: 7th International Conference on Computer Analysis of Images and Patterns, Kiel, Germany. Sep, 1997.
- cucim.skimage.filters.farid_v(image, *, mask=None)#
Find the vertical edges of an image using the Farid transform.
- Parameters
- image2-D array
Image to process.
- mask2-D array, optional
An optional mask to limit the application to a certain area. Note that pixels surrounding masked regions are also masked to prevent masked regions from affecting the result.
- Returns
- output2-D array
The Farid edge map.
Notes
The kernel was constructed using the 5-tap weights from [1].
References
- 1
Farid, H. and Simoncelli, E. P., “Differentiation of discrete multidimensional signals”, IEEE Transactions on Image Processing 13(4): 496-508, 2004. DOI:10.1109/TIP.2004.823819
- cucim.skimage.filters.frangi(image, sigmas=range(1, 10, 2), scale_range=None, scale_step=None, alpha=0.5, beta=0.5, gamma=15, black_ridges=True, mode='reflect', cval=0)#
Filter an image with the Frangi vesselness filter.
This filter can be used to detect continuous ridges, e.g. vessels, wrinkles, rivers. It can be used to calculate the fraction of the whole image containing such objects.
Defined only for 2-D and 3-D images. Calculates the eigenvectors of the Hessian to compute the similarity of an image region to vessels, according to the method described in [1].
- Parameters
- image(N, M[, P]) ndarray
Array with input image data.
- sigmasiterable of floats, optional
Sigmas used as scales of filter, i.e., np.arange(scale_range[0], scale_range[1], scale_step)
- scale_range2-tuple of floats, optional
The range of sigmas used.
- scale_stepfloat, optional
Step size between sigmas.
- alphafloat, optional
Frangi correction constant that adjusts the filter’s sensitivity to deviation from a plate-like structure.
- betafloat, optional
Frangi correction constant that adjusts the filter’s sensitivity to deviation from a blob-like structure.
- gammafloat, optional
Frangi correction constant that adjusts the filter’s sensitivity to areas of high variance/texture/structure.
- black_ridgesboolean, optional
When True (the default), the filter detects black ridges; when False, it detects white ridges.
- mode{‘constant’, ‘reflect’, ‘wrap’, ‘nearest’, ‘mirror’}, optional
How to handle values outside the image borders.
- cvalfloat, optional
Used in conjunction with mode ‘constant’, the value outside the image boundaries.
- Returns
- out(N, M[, P]) ndarray
Filtered image (maximum of pixels across all scales).
Notes
Written by Marc Schrijver, November 2001 Re-Written by D. J. Kroon, University of Twente, May 2009, [2] Adoption of 3D version from D. G. Ellis, Januar 20017, [3]
References
- 1
Frangi, A. F., Niessen, W. J., Vincken, K. L., & Viergever, M. A. (1998,). Multiscale vessel enhancement filtering. In International Conference on Medical Image Computing and Computer-Assisted Intervention (pp. 130-137). Springer Berlin Heidelberg. DOI:10.1007/BFb0056195
- 2
Kroon, D. J.: Hessian based Frangi vesselness filter.
- 3
Ellis, D. G.: https://github.com/ellisdg/frangi3d/tree/master/frangi
- cucim.skimage.filters.gabor(image, frequency, theta=0, bandwidth=1, sigma_x=None, sigma_y=None, n_stds=3, offset=0, mode='reflect', cval=0)#
Return real and imaginary responses to Gabor filter.
The real and imaginary parts of the Gabor filter kernel are applied to the image and the response is returned as a pair of arrays.
Gabor filter is a linear filter with a Gaussian kernel which is modulated by a sinusoidal plane wave. Frequency and orientation representations of the Gabor filter are similar to those of the human visual system. Gabor filter banks are commonly used in computer vision and image processing. They are especially suitable for edge detection and texture classification.
- Parameters
- image2-D array
Input image.
- frequencyfloat
Spatial frequency of the harmonic function. Specified in pixels.
- thetafloat, optional
Orientation in radians. If 0, the harmonic is in the x-direction.
- bandwidthfloat, optional
The bandwidth captured by the filter. For fixed bandwidth,
sigma_xandsigma_ywill decrease with increasing frequency. This value is ignored ifsigma_xandsigma_yare set by the user.- sigma_x, sigma_yfloat, optional
Standard deviation in x- and y-directions. These directions apply to the kernel before rotation. If theta = pi/2, then the kernel is rotated 90 degrees so that
sigma_xcontrols the vertical direction.- n_stdsscalar, optional
The linear size of the kernel is n_stds (3 by default) standard deviations.
- offsetfloat, optional
Phase offset of harmonic function in radians.
- mode{‘constant’, ‘nearest’, ‘reflect’, ‘mirror’, ‘wrap’}, optional
Mode used to convolve image with a kernel, passed to ndi.convolve
- cvalscalar, optional
Value to fill past edges of input if
modeof convolution is ‘constant’. The parameter is passed to ndi.convolve.
- Returns
- real, imagarrays
Filtered images using the real and imaginary parts of the Gabor filter kernel. Images are of the same dimensions as the input one.
References
Examples
>>> import cupy as cp >>> from cucim.skimage.filters import gabor >>> from skimage import data, io >>> from matplotlib import pyplot as plt
>>> image = cp.array(data.coins()) >>> # detecting edges in a coin image >>> filt_real, filt_imag = gabor(image, frequency=0.6) >>> plt.figure() >>> io.imshow(cp.asnumpy(filt_real)) >>> io.show()
>>> # less sensitivity to finer details with the lower frequency kernel >>> filt_real, filt_imag = gabor(image, frequency=0.1) >>> plt.figure() >>> io.imshow(cp.asnumpy(filt_real) >>> io.show()
- cucim.skimage.filters.gabor_kernel(frequency, theta=0, bandwidth=1, sigma_x=None, sigma_y=None, n_stds=3, offset=0, dtype=None, *, float_dtype=None)#
Return complex 2D Gabor filter kernel.
Gabor kernel is a Gaussian kernel modulated by a complex harmonic function. Harmonic function consists of an imaginary sine function and a real cosine function. Spatial frequency is inversely proportional to the wavelength of the harmonic and to the standard deviation of a Gaussian kernel. The bandwidth is also inversely proportional to the standard deviation.
- Parameters
- frequencyfloat
Spatial frequency of the harmonic function. Specified in pixels.
- thetafloat, optional
Orientation in radians. If 0, the harmonic is in the x-direction.
- bandwidthfloat, optional
The bandwidth captured by the filter. For fixed bandwidth,
sigma_xandsigma_ywill decrease with increasing frequency. This value is ignored ifsigma_xandsigma_yare set by the user.- sigma_x, sigma_yfloat, optional
Standard deviation in x- and y-directions. These directions apply to the kernel before rotation. If theta = pi/2, then the kernel is rotated 90 degrees so that
sigma_xcontrols the vertical direction.- n_stdsscalar, optional
The linear size of the kernel is n_stds (3 by default) standard deviations
- offsetfloat, optional
Phase offset of harmonic function in radians.
- dtype{np.complex64, np.complex128}
Specifies if the filter is single or double precision complex.
- Returns
- gcomplex array
Complex filter kernel.
References
Examples
>>> import cupy as cp >>> from cucim.skimage.filters import gabor_kernel >>> from skimage import io >>> from matplotlib import pyplot as plt
>>> gk = gabor_kernel(frequency=0.2) >>> plt.figure() >>> io.imshow(cp.asnumpy(gk.real)) >>> io.show()
>>> # more ripples (equivalent to increasing the size of the >>> # Gaussian spread) >>> gk = gabor_kernel(frequency=0.2, bandwidth=0.1) >>> plt.figure() >>> io.imshow(cp.asnumpy(gk.real)) >>> io.show()
- cucim.skimage.filters.gaussian(image, sigma=1, output=None, mode='nearest', cval=0, multichannel=None, preserve_range=False, truncate=4.0, *, channel_axis=None)#
Multi-dimensional Gaussian filter.
- Parameters
- imagearray-like
Input image (grayscale or color) to filter.
- sigmascalar or sequence of scalars, optional
Standard deviation for Gaussian kernel. The standard deviations of the Gaussian filter are given for each axis as a sequence, or as a single number, in which case it is equal for all axes.
- outputarray, optional
The
outputparameter passes an array in which to store the filter output.- mode{‘reflect’, ‘constant’, ‘nearest’, ‘mirror’, ‘wrap’}, optional
The
modeparameter determines how the array borders are handled, wherecvalis the value when mode is equal to ‘constant’. Default is ‘nearest’.- cvalscalar, optional
Value to fill past edges of input if
modeis ‘constant’. Default is 0.0- multichannelbool, optional (default: None)
Whether the last axis of the image is to be interpreted as multiple channels. If True, each channel is filtered separately (channels are not mixed together). Only 3 channels are supported. If
None, the function will attempt to guess this, and raise a warning if ambiguous, when the array has shape (M, N, 3). This argument is deprecated: specify channel_axis instead.- preserve_rangebool, optional
Whether to keep the original range of values. Otherwise, the input image is converted according to the conventions of
img_as_float. Also see https://scikit-image.org/docs/dev/user_guide/data_types.html- truncatefloat, optional
Truncate the filter at this many standard deviations.
- channel_axisint or None, optional
If None, the image is assumed to be a grayscale (single channel) image. Otherwise, this parameter indicates which axis of the array corresponds to channels.
- Returns
- filtered_imagendarray
the filtered array
- Other Parameters
- multichannelDEPRECATED
Deprecated in favor of channel_axis.
Deprecated since version 22.02.00.
Notes
This function is a wrapper around
scipy.ndi.gaussian_filter().Integer arrays are converted to float.
The
outputshould be floating point data type since gaussian converts to float providedimage. Ifoutputis not provided, another array will be allocated and returned as the result.The multi-dimensional filter is implemented as a sequence of one-dimensional convolution filters. The intermediate arrays are stored in the same data type as the output. Therefore, for output types with a limited precision, the results may be imprecise because intermediate results may be stored with insufficient precision.
Examples
>>> import cupy as cp >>> a = cp.zeros((3, 3)) >>> a[1, 1] = 1 >>> a array([[0., 0., 0.], [0., 1., 0.], [0., 0., 0.]]) >>> gaussian(a, sigma=0.4) # mild smoothing array([[0.00163116, 0.03712502, 0.00163116], [0.03712502, 0.84496158, 0.03712502], [0.00163116, 0.03712502, 0.00163116]]) >>> gaussian(a, sigma=1) # more smoothing array([[0.05855018, 0.09653293, 0.05855018], [0.09653293, 0.15915589, 0.09653293], [0.05855018, 0.09653293, 0.05855018]]) >>> # Several modes are possible for handling boundaries >>> gaussian(a, sigma=1, mode='reflect') array([[0.08767308, 0.12075024, 0.08767308], [0.12075024, 0.16630671, 0.12075024], [0.08767308, 0.12075024, 0.08767308]]) >>> # For RGB images, each is filtered separately >>> from skimage.data import astronaut >>> image = cp.array(astronaut()) >>> filtered_img = gaussian(image, sigma=1, channel_axis=-1)
- cucim.skimage.filters.hessian(image, sigmas=range(1, 10, 2), scale_range=None, scale_step=None, alpha=0.5, beta=0.5, gamma=15, black_ridges=True, mode='reflect', cval=0)#
Filter an image with the Hybrid Hessian filter.
This filter can be used to detect continuous edges, e.g. vessels, wrinkles, rivers. It can be used to calculate the fraction of the whole image containing such objects.
Defined only for 2-D and 3-D images. Almost equal to Frangi filter, but uses alternative method of smoothing. Refer to [1] to find the differences between Frangi and Hessian filters.
- Parameters
- image(N, M[, P]) ndarray
Array with input image data.
- sigmasiterable of floats, optional
Sigmas used as scales of filter, i.e., np.arange(scale_range[0], scale_range[1], scale_step)
- scale_range2-tuple of floats, optional
The range of sigmas used.
- scale_stepfloat, optional
Step size between sigmas.
- betafloat, optional
Frangi correction constant that adjusts the filter’s sensitivity to deviation from a blob-like structure.
- gammafloat, optional
Frangi correction constant that adjusts the filter’s sensitivity to areas of high variance/texture/structure.
- black_ridgesboolean, optional
When True (the default), the filter detects black ridges; when False, it detects white ridges.
- mode{‘constant’, ‘reflect’, ‘wrap’, ‘nearest’, ‘mirror’}, optional
How to handle values outside the image borders.
- cvalfloat, optional
Used in conjunction with mode ‘constant’, the value outside the image boundaries.
- Returns
- out(N, M[, P]) ndarray
Filtered image (maximum of pixels across all scales).
Notes
Written by Marc Schrijver (November 2001) Re-Written by D. J. Kroon University of Twente (May 2009) [2]
References
- 1
Ng, C. C., Yap, M. H., Costen, N., & Li, B. (2014,). Automatic wrinkle detection using hybrid Hessian filter. In Asian Conference on Computer Vision (pp. 609-622). Springer International Publishing. DOI:10.1007/978-3-319-16811-1_40
- 2
Kroon, D. J.: Hessian based Frangi vesselness filter.
- cucim.skimage.filters.inverse(data, impulse_response=None, filter_params={}, max_gain=2, predefined_filter=None)#
Apply the filter in reverse to the given data.
- Parameters
- data(M,N) ndarray
Input data.
- impulse_responsecallable f(r, c, **filter_params)
Impulse response of the filter. See LPIFilter2D.__init__.
- filter_paramsdict
Additional keyword parameters to the impulse_response function.
- max_gainfloat
Limit the filter gain. Often, the filter contains zeros, which would cause the inverse filter to have infinite gain. High gain causes amplification of artefacts, so a conservative limit is recommended.
- Other Parameters
- predefined_filterLPIFilter2D
If you need to apply the same filter multiple times over different images, construct the LPIFilter2D and specify it here.
- cucim.skimage.filters.laplace(image, ksize=3, mask=None)#
Find the edges of an image using the Laplace operator.
- Parameters
- imagendarray
Image to process.
- ksizeint, optional
Define the size of the discrete Laplacian operator such that it will have a size of (ksize,) * image.ndim.
- maskndarray, optional
An optional mask to limit the application to a certain area. Note that pixels surrounding masked regions are also masked to prevent masked regions from affecting the result.
- Returns
- outputndarray
The Laplace edge map.
Notes
The Laplacian operator is generated using the function skimage.restoration.uft.laplacian().
- cucim.skimage.filters.median(image, footprint=None, out=None, mode='nearest', cval=0.0, behavior='ndimage', *, algorithm='auto', algorithm_kwargs={})#
Return local median of an image.
- Parameters
- imagearray-like
Input image.
- footprintndarray, optional
If
behavior=='rank',footprintis a 2-D array of 1’s and 0’s. Ifbehavior=='ndimage',footprintis a N-D array of 1’s and 0’s with the same number of dimension thanimage. If None,footprintwill be a N-D array with 3 elements for each dimension (e.g., vector, square, cube, etc.)- outndarray, (same dtype as image), optional
If None, a new array is allocated.
- mode{‘reflect’, ‘constant’, ‘nearest’, ‘mirror’,’‘wrap’}, optional
The mode parameter determines how the array borders are handled, where
cvalis the value when mode is equal to ‘constant’. Default is ‘nearest’.New in version 0.15:
modeis used whenbehavior='ndimage'.- cvalscalar, optional
Value to fill past edges of input if mode is ‘constant’. Default is 0.0
New in version 0.15:
cvalwas added in 0.15 is used whenbehavior='ndimage'.- behavior{‘ndimage’, ‘rank’}, optional
Either to use the old behavior (i.e., < 0.15) or the new behavior. The old behavior will call the
skimage.filters.rank.median(). The new behavior will call thescipy.ndimage.median_filter(). Default is ‘ndimage’.New in version 0.15:
behavioris introduced in 0.15Changed in version 0.16: Default
behaviorhas been changed from ‘rank’ to ‘ndimage’
- Returns
- out2-D array (same dtype as input image)
Output image.
- Other Parameters
- algorithm{‘auto’, ‘histogram’, ‘sorting’}
Determines which algorithm is used to compute the median. The default of ‘auto’ will attempt to use a histogram-based algorithm for 2D images with 8 or 16-bit integer data types. Otherwise a sorting-based algorithm will be used. Note: this paramter is cuCIM-specific and does not exist in upstream scikit-image.
- algorithm_kwargsdict
Any additional algorithm-specific keywords. Currently can only be used to set the number of parallel partitions for the ‘histogram’ algorithm. (e.g.
algorithm_kwargs={'partitions': 256}). Note: this paramter is cuCIM-specific and does not exist in upstream scikit-image.- selemDEPRECATED
Deprecated in favor of footprint.
Deprecated since version 22.02.00.
See also
skimage.filters.rank.median()Rank-based implementation of the median filtering offering more flexibility with additional parameters but dedicated for unsigned integer images.
Notes
An efficient, histogram-based median filter as described in [1] is faster than the sorting based approach for larger kernel sizes (e.g. greater than 13x13 or so in 2D). It has near-constant run time regardless of the kernel size. The algorithm presented in [1] has been adapted to additional bit depths here. When algorithm=’auto’, the histogram-based algorithm will be chosen for integer-valued images with sufficiently large footprint size. Otherwise, the sorting-based approach is used.
References
- 1(1,2)
O. Green, “Efficient Scalable Median Filtering Using Histogram-Based Operations,” in IEEE Transactions on Image Processing, vol. 27, no. 5, pp. 2217-2228, May 2018, https://doi.org/10.1109/TIP.2017.2781375.
Examples
>>> import cupy as cp >>> from skimage import data >>> from cucim.skimage.morphology import disk >>> from cucim.skimage.filters import median >>> img = cp.array(data.camera()) >>> med = median(img, disk(5))
- cucim.skimage.filters.meijering(image, sigmas=range(1, 10, 2), alpha=None, black_ridges=True, mode='reflect', cval=0)#
Filter an image with the Meijering neuriteness filter.
This filter can be used to detect continuous ridges, e.g. neurites, wrinkles, rivers. It can be used to calculate the fraction of the whole image containing such objects.
Calculates the eigenvectors of the Hessian to compute the similarity of an image region to neurites, according to the method described in [1].
- Parameters
- image(N, M[, …, P]) ndarray
Array with input image data.
- sigmasiterable of floats, optional
Sigmas used as scales of filter
- alphafloat, optional
Frangi correction constant that adjusts the filter’s sensitivity to deviation from a plate-like structure.
- black_ridgesboolean, optional
When True (the default), the filter detects black ridges; when False, it detects white ridges.
- mode{‘constant’, ‘reflect’, ‘wrap’, ‘nearest’, ‘mirror’}, optional
How to handle values outside the image borders.
- cvalfloat, optional
Used in conjunction with mode ‘constant’, the value outside the image boundaries.
- Returns
- out(N, M[, …, P]) ndarray
Filtered image (maximum of pixels across all scales).
References
- 1
Meijering, E., Jacob, M., Sarria, J. C., Steiner, P., Hirling, H., Unser, M. (2004). Design and validation of a tool for neurite tracing and analysis in fluorescence microscopy images. Cytometry Part A, 58(2), 167-176. DOI:10.1002/cyto.a.20022
- cucim.skimage.filters.prewitt(image, mask=None, *, axis=None, mode='reflect', cval=0.0)#
Find the edge magnitude using the Prewitt transform.
- Parameters
- imagearray
The input image.
- maskarray of bool, optional
Clip the output image to this mask. (Values where mask=0 will be set to 0.)
- axisint or sequence of int, optional
Compute the edge filter along this axis. If not provided, the edge magnitude is computed. This is defined as:
prw_mag = np.sqrt(sum([prewitt(image, axis=i)**2 for i in range(image.ndim)]) / image.ndim)
The magnitude is also computed if axis is a sequence.
- modestr or sequence of str, optional
The boundary mode for the convolution. See scipy.ndimage.convolve for a description of the modes. This can be either a single boundary mode or one boundary mode per axis.
- cvalfloat, optional
When mode is
'constant', this is the constant used in values outside the boundary of the image data.
- Returns
- outputarray of float
The Prewitt edge map.
See also
Notes
The edge magnitude depends slightly on edge directions, since the approximation of the gradient operator by the Prewitt operator is not completely rotation invariant. For a better rotation invariance, the Scharr operator should be used. The Sobel operator has a better rotation invariance than the Prewitt operator, but a worse rotation invariance than the Scharr operator.
Examples
>>> import cupy as cp >>> from skimage import data >>> from cucim.skimage import filters >>> camera = cp.array(data.camera()) >>> edges = filters.prewitt(camera)
- cucim.skimage.filters.prewitt_h(image, mask=None)#
Find the horizontal edges of an image using the Prewitt transform.
- Parameters
- image2-D array
Image to process.
- mask2-D array, optional
An optional mask to limit the application to a certain area. Note that pixels surrounding masked regions are also masked to prevent masked regions from affecting the result.
- Returns
- output2-D array
The Prewitt edge map.
Notes
We use the following kernel:
1/3 1/3 1/3 0 0 0 -1/3 -1/3 -1/3
- cucim.skimage.filters.prewitt_v(image, mask=None)#
Find the vertical edges of an image using the Prewitt transform.
- Parameters
- image2-D array
Image to process.
- mask2-D array, optional
An optional mask to limit the application to a certain area. Note that pixels surrounding masked regions are also masked to prevent masked regions from affecting the result.
- Returns
- output2-D array
The Prewitt edge map.
Notes
We use the following kernel:
1/3 0 -1/3 1/3 0 -1/3 1/3 0 -1/3
- cucim.skimage.filters.rank_order(image)#
Return an image of the same shape where each pixel is the index of the pixel value in the ascending order of the unique values of
image, aka the rank-order value.- Parameters
- imagendarray
- Returns
- labelsndarray of type np.uint32, of shape image.shape
New array where each pixel has the rank-order value of the corresponding pixel in
image. Pixel values are between 0 and n - 1, where n is the number of distinct unique values inimage.- original_values1-D ndarray
Unique original values of
image
Examples
>>> a = cp.asarray([[1, 4, 5], [4, 4, 1], [5, 1, 1]]) >>> a array([[1, 4, 5], [4, 4, 1], [5, 1, 1]]) >>> rank_order(a) (array([[0, 1, 2], [1, 1, 0], [2, 0, 0]], dtype=uint32), array([1, 4, 5])) >>> b = cp.asarray([-1., 2.5, 3.1, 2.5]) >>> rank_order(b) (array([0, 1, 2, 1], dtype=uint32), array([-1. , 2.5, 3.1]))
- cucim.skimage.filters.roberts(image, mask=None)#
Find the edge magnitude using Roberts’ cross operator.
- Parameters
- image2-D array
Image to process.
- mask2-D array, optional
An optional mask to limit the application to a certain area. Note that pixels surrounding masked regions are also masked to prevent masked regions from affecting the result.
- Returns
- output2-D array
The Roberts’ Cross edge map.
See also
roberts_pos_diag,roberts_neg_diagdiagonal edge detection.
sobel,scharr,prewitt,cucim.skimage.feature.canny
Examples
>>> import cupy as cp >>> from skimage import data >>> camera = cp.array(data.camera()) >>> from cucim.skimage import filters >>> edges = filters.roberts(camera)
- cucim.skimage.filters.roberts_neg_diag(image, mask=None)#
Find the cross edges of an image using the Roberts’ Cross operator.
The kernel is applied to the input image to produce separate measurements of the gradient component one orientation.
- Parameters
- image2-D array
Image to process.
- mask2-D array, optional
An optional mask to limit the application to a certain area. Note that pixels surrounding masked regions are also masked to prevent masked regions from affecting the result.
- Returns
- output2-D array
The Robert’s edge map.
Notes
We use the following kernel:
0 1 -1 0
- cucim.skimage.filters.roberts_pos_diag(image, mask=None)#
Find the cross edges of an image using Roberts’ cross operator.
The kernel is applied to the input image to produce separate measurements of the gradient component one orientation.
- Parameters
- image2-D array
Image to process.
- mask2-D array, optional
An optional mask to limit the application to a certain area. Note that pixels surrounding masked regions are also masked to prevent masked regions from affecting the result.
- Returns
- output2-D array
The Robert’s edge map.
Notes
We use the following kernel:
1 0 0 -1
- cucim.skimage.filters.sato(image, sigmas=range(1, 10, 2), black_ridges=True, mode='reflect', cval=0)#
Filter an image with the Sato tubeness filter.
This filter can be used to detect continuous ridges, e.g. tubes, wrinkles, rivers. It can be used to calculate the fraction of the whole image containing such objects.
Defined only for 2-D and 3-D images. Calculates the eigenvectors of the Hessian to compute the similarity of an image region to tubes, according to the method described in [1].
- Parameters
- image(N, M[, P]) ndarray
Array with input image data.
- sigmasiterable of floats, optional
Sigmas used as scales of filter.
- black_ridgesboolean, optional
When True (the default), the filter detects black ridges; when False, it detects white ridges.
- mode{‘constant’, ‘reflect’, ‘wrap’, ‘nearest’, ‘mirror’}, optional
How to handle values outside the image borders.
- cvalfloat, optional
Used in conjunction with mode ‘constant’, the value outside the image boundaries.
- Returns
- out(N, M[, P]) ndarray
Filtered image (maximum of pixels across all scales).
References
- 1
Sato, Y., Nakajima, S., Shiraga, N., Atsumi, H., Yoshida, S., Koller, T., …, Kikinis, R. (1998). Three-dimensional multi-scale line filter for segmentation and visualization of curvilinear structures in medical images. Medical image analysis, 2(2), 143-168. DOI:10.1016/S1361-8415(98)80009-1
- cucim.skimage.filters.scharr(image, mask=None, *, axis=None, mode='reflect', cval=0.0)#
Find the edge magnitude using the Scharr transform.
- Parameters
- imagearray
The input image.
- maskarray of bool, optional
Clip the output image to this mask. (Values where mask=0 will be set to 0.)
- axisint or sequence of int, optional
Compute the edge filter along this axis. If not provided, the edge magnitude is computed. This is defined as:
sch_mag = np.sqrt(sum([scharr(image, axis=i)**2 for i in range(image.ndim)]) / image.ndim)
The magnitude is also computed if axis is a sequence.
- modestr or sequence of str, optional
The boundary mode for the convolution. See scipy.ndimage.convolve for a description of the modes. This can be either a single boundary mode or one boundary mode per axis.
- cvalfloat, optional
When mode is
'constant', this is the constant used in values outside the boundary of the image data.
- Returns
- outputarray of float
The Scharr edge map.
See also
Notes
The Scharr operator has a better rotation invariance than other edge filters such as the Sobel or the Prewitt operators.
References
- 1
D. Kroon, 2009, Short Paper University Twente, Numerical Optimization of Kernel Based Image Derivatives.
- 2
https://en.wikipedia.org/wiki/Sobel_operator#Alternative_operators
Examples
>>> import cupy as cp >>> from skimage import data >>> from cucim.skimage import filters >>> camera = cp.array(data.camera()) >>> edges = filters.scharr(camera)
- cucim.skimage.filters.scharr_h(image, mask=None)#
Find the horizontal edges of an image using the Scharr transform.
- Parameters
- image2-D array
Image to process.
- mask2-D array, optional
An optional mask to limit the application to a certain area. Note that pixels surrounding masked regions are also masked to prevent masked regions from affecting the result.
- Returns
- output2-D array
The Scharr edge map.
Notes
We use the following kernel:
3 10 3 0 0 0 -3 -10 -3
References
- 1
D. Kroon, 2009, Short Paper University Twente, Numerical Optimization of Kernel Based Image Derivatives.
- cucim.skimage.filters.scharr_v(image, mask=None)#
Find the vertical edges of an image using the Scharr transform.
- Parameters
- image2-D array
Image to process
- mask2-D array, optional
An optional mask to limit the application to a certain area. Note that pixels surrounding masked regions are also masked to prevent masked regions from affecting the result.
- Returns
- output2-D array
The Scharr edge map.
Notes
We use the following kernel:
3 0 -3 10 0 -10 3 0 -3
References
- 1
D. Kroon, 2009, Short Paper University Twente, Numerical Optimization of Kernel Based Image Derivatives.
- cucim.skimage.filters.sobel(image, mask=None, *, axis=None, mode='reflect', cval=0.0)#
Find edges in an image using the Sobel filter.
- Parameters
- imagearray
The input image.
- maskarray of bool, optional
Clip the output image to this mask. (Values where mask=0 will be set to 0.)
- axisint or sequence of int, optional
Compute the edge filter along this axis. If not provided, the edge magnitude is computed. This is defined as:
sobel_mag = np.sqrt(sum([sobel(image, axis=i)**2 for i in range(image.ndim)]) / image.ndim)
The magnitude is also computed if axis is a sequence.
- modestr or sequence of str, optional
The boundary mode for the convolution. See scipy.ndimage.convolve for a description of the modes. This can be either a single boundary mode or one boundary mode per axis.
- cvalfloat, optional
When mode is
'constant', this is the constant used in values outside the boundary of the image data.
- Returns
- outputarray of float
The Sobel edge map.
See also
References
- 1
D. Kroon, 2009, Short Paper University Twente, Numerical Optimization of Kernel Based Image Derivatives.
- 2
Examples
>>> import cupy as cp >>> from skimage import data >>> from cucim.skimage import filters >>> camera = cp.array(data.camera()) >>> edges = filters.sobel(camera)
- cucim.skimage.filters.sobel_h(image, mask=None)#
Find the horizontal edges of an image using the Sobel transform.
- Parameters
- image2-D array
Image to process.
- mask2-D array, optional
An optional mask to limit the application to a certain area. Note that pixels surrounding masked regions are also masked to prevent masked regions from affecting the result.
- Returns
- output2-D array
The Sobel edge map.
Notes
We use the following kernel:
1 2 1 0 0 0 -1 -2 -1
- cucim.skimage.filters.sobel_v(image, mask=None)#
Find the vertical edges of an image using the Sobel transform.
- Parameters
- image2-D array
Image to process.
- mask2-D array, optional
An optional mask to limit the application to a certain area. Note that pixels surrounding masked regions are also masked to prevent masked regions from affecting the result.
- Returns
- output2-D array
The Sobel edge map.
Notes
We use the following kernel:
1 0 -1 2 0 -2 1 0 -1
- cucim.skimage.filters.threshold_isodata(image=None, nbins=256, return_all=False, *, hist=None)#
Return threshold value(s) based on ISODATA method.
Histogram-based threshold, known as Ridler-Calvard method or inter-means. Threshold values returned satisfy the following equality:
threshold = (image[image <= threshold].mean() + image[image > threshold].mean()) / 2.0
That is, returned thresholds are intensities that separate the image into two groups of pixels, where the threshold intensity is midway between the mean intensities of these groups.
For integer images, the above equality holds to within one; for floating- point images, the equality holds to within the histogram bin-width.
Either image or hist must be provided. In case hist is given, the actual histogram of the image is ignored.
- Parameters
- image(N, M[, …, P]) ndarray
Grayscale input image.
- nbinsint, optional
Number of bins used to calculate histogram. This value is ignored for integer arrays.
- return_allbool, optional
If False (default), return only the lowest threshold that satisfies the above equality. If True, return all valid thresholds.
- histarray, or 2-tuple of arrays, optional
Histogram to determine the threshold from and a corresponding array of bin center intensities. Alternatively, only the histogram can be passed.
- Returns
- thresholdfloat or int or array
Threshold value(s).
References
- 1
Ridler, TW & Calvard, S (1978), “Picture thresholding using an iterative selection method” IEEE Transactions on Systems, Man and Cybernetics 8: 630-632, DOI:10.1109/TSMC.1978.4310039
- 2
Sezgin M. and Sankur B. (2004) “Survey over Image Thresholding Techniques and Quantitative Performance Evaluation” Journal of Electronic Imaging, 13(1): 146-165, http://www.busim.ee.boun.edu.tr/~sankur/SankurFolder/Threshold_survey.pdf DOI:10.1117/1.1631315
- 3
ImageJ AutoThresholder code, http://fiji.sc/wiki/index.php/Auto_Threshold
Examples
>>> from skimage.data import coins >>> image = cp.array(coins()) >>> thresh = threshold_isodata(image) >>> binary = image > thresh
- cucim.skimage.filters.threshold_li(image, *, tolerance=None, initial_guess=None, iter_callback=None)#
Compute threshold value by Li’s iterative Minimum Cross Entropy method.
- Parameters
- image(N, M[, …, P]) ndarray
Grayscale input image.
- tolerancefloat, optional
Finish the computation when the change in the threshold in an iteration is less than this value. By default, this is half the smallest difference between intensity values in
image.- initial_guessfloat or Callable[[array[float]], float], optional
Li’s iterative method uses gradient descent to find the optimal threshold. If the image intensity histogram contains more than two modes (peaks), the gradient descent could get stuck in a local optimum. An initial guess for the iteration can help the algorithm find the globally-optimal threshold. A float value defines a specific start point, while a callable should take in an array of image intensities and return a float value. Example valid callables include
numpy.mean(default),lambda arr: numpy.quantile(arr, 0.95), or evenskimage.filters.threshold_otsu().- iter_callbackCallable[[float], Any], optional
A function that will be called on the threshold at every iteration of the algorithm.
- Returns
- thresholdfloat
Upper threshold value. All pixels with an intensity higher than this value are assumed to be foreground.
References
- 1
Li C.H. and Lee C.K. (1993) “Minimum Cross Entropy Thresholding” Pattern Recognition, 26(4): 617-625 DOI:10.1016/0031-3203(93)90115-D
- 2
Li C.H. and Tam P.K.S. (1998) “An Iterative Algorithm for Minimum Cross Entropy Thresholding” Pattern Recognition Letters, 18(8): 771-776 DOI:10.1016/S0167-8655(98)00057-9
- 3
Sezgin M. and Sankur B. (2004) “Survey over Image Thresholding Techniques and Quantitative Performance Evaluation” Journal of Electronic Imaging, 13(1): 146-165 DOI:10.1117/1.1631315
- 4
ImageJ AutoThresholder code, http://fiji.sc/wiki/index.php/Auto_Threshold
Examples
>>> from skimage.data import camera >>> image = cp.array(camera()) >>> thresh = threshold_li(image) >>> binary = image > thresh
- cucim.skimage.filters.threshold_local(image, block_size=3, method='gaussian', offset=0, mode='reflect', param=None, cval=0)#
Compute a threshold mask image based on local pixel neighborhood.
Also known as adaptive or dynamic thresholding. The threshold value is the weighted mean for the local neighborhood of a pixel subtracted by a constant. Alternatively the threshold can be determined dynamically by a given function, using the ‘generic’ method.
- Parameters
- image(N, M[, …, P]) ndarray
Grayscale input image.
- block_sizeint or sequence of int
Odd size of pixel neighborhood which is used to calculate the threshold value (e.g. 3, 5, 7, …, 21, …).
- method{‘generic’, ‘gaussian’, ‘mean’, ‘median’}, optional
Method used to determine adaptive threshold for local neighbourhood in weighted mean image.
‘generic’: use custom function (see
paramparameter)‘gaussian’: apply gaussian filter (see
paramparameter for custom sigma value)‘mean’: apply arithmetic mean filter
‘median’: apply median rank filter
By default the ‘gaussian’ method is used.
- offsetfloat, optional
Constant subtracted from weighted mean of neighborhood to calculate the local threshold value. Default offset is 0.
- mode{‘reflect’, ‘constant’, ‘nearest’, ‘mirror’, ‘wrap’}, optional
The mode parameter determines how the array borders are handled, where cval is the value when mode is equal to ‘constant’. Default is ‘reflect’.
- param{int, function}, optional
Either specify sigma for ‘gaussian’ method or function object for ‘generic’ method. This functions takes the flat array of local neighbourhood as a single argument and returns the calculated threshold for the centre pixel.
- cvalfloat, optional
Value to fill past edges of input if mode is ‘constant’.
- Returns
- threshold(N, M[, …, P]) ndarray
Threshold image. All pixels in the input image higher than the corresponding pixel in the threshold image are considered foreground.
References
- 1
Gonzalez, R. C. and Wood, R. E. “Digital Image Processing (2nd Edition).” Prentice-Hall Inc., 2002: 600–612. ISBN: 0-201-18075-8
Examples
>>> import cupy as cp >>> from skimage.data import camera >>> image = cp.array(camera()[:50, :50]) >>> binary_image1 = image > threshold_local(image, 15, 'mean')
- cucim.skimage.filters.threshold_mean(image)#
Return threshold value based on the mean of grayscale values.
- Parameters
- image(N, M[, …, P]) ndarray
Grayscale input image.
- Returns
- thresholdfloat
Upper threshold value. All pixels with an intensity higher than this value are assumed to be foreground.
References
- 1
C. A. Glasbey, “An analysis of histogram-based thresholding algorithms,” CVGIP: Graphical Models and Image Processing, vol. 55, pp. 532-537, 1993. DOI:10.1006/cgip.1993.1040
Examples
>>> from skimage.data import camera >>> image = cp.array(camera()) >>> thresh = threshold_mean(image) >>> binary = image > thresh
- cucim.skimage.filters.threshold_minimum(image=None, nbins=256, max_num_iter=10000, *, hist=None)#
Return threshold value based on minimum method.
The histogram of the input
imageis computed if not provided and smoothed until there are only two maxima. Then the minimum in between is the threshold value.Either image or hist must be provided. In case hist is given, the actual histogram of the image is ignored.
- Parameters
- image(N, M[, …, P]) ndarray, optional
Grayscale input image.
- nbinsint, optional
Number of bins used to calculate histogram. This value is ignored for integer arrays.
- max_num_iterint, optional
Maximum number of iterations to smooth the histogram.
- histarray, or 2-tuple of arrays, optional
Histogram to determine the threshold from and a corresponding array of bin center intensities. Alternatively, only the histogram can be passed.
- Returns
- thresholdfloat
Upper threshold value. All pixels with an intensity higher than this value are assumed to be foreground.
- Other Parameters
- max_iterDEPRECATED
Deprecated in favor of max_num_iter.
Deprecated since version 22.02.00.
- Raises
- RuntimeError
If unable to find two local maxima in the histogram or if the smoothing takes more than 1e4 iterations.
References
- 1
C. A. Glasbey, “An analysis of histogram-based thresholding algorithms,” CVGIP: Graphical Models and Image Processing, vol. 55, pp. 532-537, 1993.
- 2
Prewitt, JMS & Mendelsohn, ML (1966), “The analysis of cell images”, Annals of the New York Academy of Sciences 128: 1035-1053 DOI:10.1111/j.1749-6632.1965.tb11715.x
Examples
>>> from skimage.data import camera >>> image = cp.array(camera()) >>> thresh = threshold_minimum(image) >>> binary = image > thresh
- cucim.skimage.filters.threshold_multiotsu(image=None, classes=3, nbins=256, *, hist=None)#
Generate classes-1 threshold values to divide gray levels in image, following Otsu’s method for multiple classes.
The threshold values are chosen to maximize the total sum of pairwise variances between the thresholded graylevel classes. See Notes and [1] for more details.
Either image or hist must be provided. If hist is provided, the actual histogram of the image is ignored.
- Parameters
- image(N, M[, …, P]) ndarray, optional
Grayscale input image.
- classesint, optional
Number of classes to be thresholded, i.e. the number of resulting regions.
- nbinsint, optional
Number of bins used to calculate the histogram. This value is ignored for integer arrays.
- histarray, or 2-tuple of arrays, optional
Histogram from which to determine the threshold, and optionally a corresponding array of bin center intensities. If no hist provided, this function will compute it from the image (see notes).
- Returns
- thresharray
Array containing the threshold values for the desired classes.
- Raises
- ValueError
If
imagecontains less grayscale value then the desired number of classes.
Notes
This implementation relies on a Cython function whose complexity is \(O\left(\frac{Ch^{C-1}}{(C-1)!}\right)\), where \(h\) is the number of histogram bins and \(C\) is the number of classes desired.
If no hist is given, this function will make use of skimage.exposure.histogram, which behaves differently than np.histogram. While both allowed, use the former for consistent behaviour.
The input image must be grayscale.
References
- 1
Liao, P-S., Chen, T-S. and Chung, P-C., “A fast algorithm for multilevel thresholding”, Journal of Information Science and Engineering 17 (5): 713-727, 2001. Available at: <https://ftp.iis.sinica.edu.tw/JISE/2001/200109_01.pdf> DOI:10.6688/JISE.2001.17.5.1
- 2
Tosa, Y., “Multi-Otsu Threshold”, a java plugin for ImageJ. Available at: <http://imagej.net/plugins/download/Multi_OtsuThreshold.java>
Examples
>>> import cupy as cp >>> from cucim.skimage.color import label2rgb >>> from skimage import data >>> image = cp.asarray(data.camera()) >>> thresholds = threshold_multiotsu(image) >>> regions = cp.digitize(image, bins=thresholds) >>> regions_colorized = label2rgb(regions)
- cucim.skimage.filters.threshold_niblack(image, window_size=15, k=0.2)#
Applies Niblack local threshold to an array.
A threshold T is calculated for every pixel in the image using the following formula:
T = m(x,y) - k * s(x,y)
where m(x,y) and s(x,y) are the mean and standard deviation of pixel (x,y) neighborhood defined by a rectangular window with size w times w centered around the pixel. k is a configurable parameter that weights the effect of standard deviation.
- Parameters
- image(N, M[, …, P]) ndarray
Grayscale input image.
- window_sizeint, or iterable of int, optional
Window size specified as a single odd integer (3, 5, 7, …), or an iterable of length
image.ndimcontaining only odd integers (e.g.(1, 5, 5)).- kfloat, optional
Value of parameter k in threshold formula.
- Returns
- threshold(N, M) ndarray
Threshold mask. All pixels with an intensity higher than this value are assumed to be foreground.
Notes
This algorithm is originally designed for text recognition.
The Bradley threshold is a particular case of the Niblack one, being equivalent to
>>> from skimage import data >>> image = cp.array(data.page()) >>> q = 1 >>> threshold_image = threshold_niblack(image, k=0) * q
for some value
q. By default, Bradley and Roth useq=1.References
- 1
W. Niblack, An introduction to Digital Image Processing, Prentice-Hall, 1986.
- 2
D. Bradley and G. Roth, “Adaptive thresholding using Integral Image”, Journal of Graphics Tools 12(2), pp. 13-21, 2007. DOI:10.1080/2151237X.2007.10129236
Examples
>>> from skimage import data >>> image = cp.array(data.page()) >>> threshold_image = threshold_niblack(image, window_size=7, k=0.1)
- cucim.skimage.filters.threshold_otsu(image=None, nbins=256, *, hist=None)#
Return threshold value based on Otsu’s method.
Either image or hist must be provided. If hist is provided, the actual histogram of the image is ignored.
- Parameters
- image(N, M[, …, P]) ndarray, optional
Grayscale input image.
- nbinsint, optional
Number of bins used to calculate histogram. This value is ignored for integer arrays.
- histarray, or 2-tuple of arrays, optional
Histogram from which to determine the threshold, and optionally a corresponding array of bin center intensities. If no hist provided, this function will compute it from the image.
- Returns
- thresholdfloat
Upper threshold value. All pixels with an intensity higher than this value are assumed to be foreground.
Notes
The input image must be grayscale.
References
- 1
Wikipedia, https://en.wikipedia.org/wiki/Otsu’s_Method
Examples
>>> from skimage.data import camera >>> image = cp.array(camera()) >>> thresh = threshold_otsu(image) >>> binary = image <= thresh
- cucim.skimage.filters.threshold_sauvola(image, window_size=15, k=0.2, r=None)#
Applies Sauvola local threshold to an array. Sauvola is a modification of Niblack technique.
In the original method a threshold T is calculated for every pixel in the image using the following formula:
T = m(x,y) * (1 + k * ((s(x,y) / R) - 1))
where m(x,y) and s(x,y) are the mean and standard deviation of pixel (x,y) neighborhood defined by a rectangular window with size w times w centered around the pixel. k is a configurable parameter that weights the effect of standard deviation. R is the maximum standard deviation of a grayscale image.
- Parameters
- image(N, M[, …, P]) ndarray
Grayscale input image.
- window_sizeint, or iterable of int, optional
Window size specified as a single odd integer (3, 5, 7, …), or an iterable of length
image.ndimcontaining only odd integers (e.g.(1, 5, 5)).- kfloat, optional
Value of the positive parameter k.
- rfloat, optional
Value of R, the dynamic range of standard deviation. If None, set to the half of the image dtype range.
- Returns
- threshold(N, M) ndarray
Threshold mask. All pixels with an intensity higher than this value are assumed to be foreground.
Notes
This algorithm is originally designed for text recognition.
References
- 1
J. Sauvola and M. Pietikainen, “Adaptive document image binarization,” Pattern Recognition 33(2), pp. 225-236, 2000. DOI:10.1016/S0031-3203(99)00055-2
Examples
>>> from skimage import data >>> image = cp.array(data.page()) >>> t_sauvola = threshold_sauvola(image, window_size=15, k=0.2) >>> binary_image = image > t_sauvola
- cucim.skimage.filters.threshold_triangle(image, nbins=256)#
Return threshold value based on the triangle algorithm.
- Parameters
- image(N, M[, …, P]) ndarray
Grayscale input image.
- nbinsint, optional
Number of bins used to calculate histogram. This value is ignored for integer arrays.
- Returns
- thresholdfloat
Upper threshold value. All pixels with an intensity higher than this value are assumed to be foreground.
References
- 1
Zack, G. W., Rogers, W. E. and Latt, S. A., 1977, Automatic Measurement of Sister Chromatid Exchange Frequency, Journal of Histochemistry and Cytochemistry 25 (7), pp. 741-753 DOI:10.1177/25.7.70454
- 2
ImageJ AutoThresholder code, http://fiji.sc/wiki/index.php/Auto_Threshold
Examples
>>> from skimage.data import camera >>> image = cp.array(camera()) >>> thresh = threshold_triangle(image) >>> binary = image > thresh
- cucim.skimage.filters.threshold_yen(image=None, nbins=256, *, hist=None)#
Return threshold value based on Yen’s method. Either image or hist must be provided. In case hist is given, the actual histogram of the image is ignored.
- Parameters
- image(N, M[, …, P]) ndarray
Grayscale input image.
- nbinsint, optional
Number of bins used to calculate histogram. This value is ignored for integer arrays.
- histarray, or 2-tuple of arrays, optional
Histogram from which to determine the threshold, and optionally a corresponding array of bin center intensities. An alternative use of this function is to pass it only hist.
- Returns
- thresholdfloat
Upper threshold value. All pixels with an intensity higher than this value are assumed to be foreground.
References
- 1
Yen J.C., Chang F.J., and Chang S. (1995) “A New Criterion for Automatic Multilevel Thresholding” IEEE Trans. on Image Processing, 4(3): 370-378. DOI:10.1109/83.366472
- 2
Sezgin M. and Sankur B. (2004) “Survey over Image Thresholding Techniques and Quantitative Performance Evaluation” Journal of Electronic Imaging, 13(1): 146-165, DOI:10.1117/1.1631315 http://www.busim.ee.boun.edu.tr/~sankur/SankurFolder/Threshold_survey.pdf
- 3
ImageJ AutoThresholder code, http://fiji.sc/wiki/index.php/Auto_Threshold
Examples
>>> from skimage.data import camera >>> image = cp.array(camera()) >>> thresh = threshold_yen(image) >>> binary = image <= thresh
- cucim.skimage.filters.try_all_threshold(image, figsize=(8, 5), verbose=True)#
Returns a figure comparing the outputs of different thresholding methods.
- Parameters
- image(N, M) ndarray
Input image.
- figsizetuple, optional
Figure size (in inches).
- verbosebool, optional
Print function name for each method.
- Returns
- fig, axtuple
Matplotlib figure and axes.
Notes
The following algorithms are used:
isodata
li
mean
minimum
otsu
triangle
yen
Examples
>>> from skimage.data import text >>> text_img = cp.array(text()) >>> fig, ax = try_all_threshold(text_img, figsize=(10, 6), verbose=False)
- cucim.skimage.filters.unsharp_mask(image, radius=1.0, amount=1.0, multichannel=False, preserve_range=False, *, channel_axis=None)#
Unsharp masking filter.
The sharp details are identified as the difference between the original image and its blurred version. These details are then scaled, and added back to the original image.
- Parameters
- image[P, …, ]M[, N][, C] ndarray
Input image.
- radiusscalar or sequence of scalars, optional
If a scalar is given, then its value is used for all dimensions. If sequence is given, then there must be exactly one radius for each dimension except the last dimension for multichannel images. Note that 0 radius means no blurring, and negative values are not allowed.
- amountscalar, optional
The details will be amplified with this factor. The factor could be 0 or negative. Typically, it is a small positive number, e.g. 1.0.
- multichannelbool, optional
If True, the last
imagedimension is considered as a color channel, otherwise as spatial. Color channels are processed individually. This argument is deprecated: specify channel_axis instead.- preserve_rangebool, optional
Whether to keep the original range of values. Otherwise, the input image is converted according to the conventions of
img_as_float. Also see https://scikit-image.org/docs/dev/user_guide/data_types.html- channel_axisint or None, optional
If None, the image is assumed to be a grayscale (single channel) image. Otherwise, this parameter indicates which axis of the array corresponds to channels.
- Returns
- output[P, …, ]M[, N][, C] ndarray of float
Image with unsharp mask applied.
- Other Parameters
- multichannelDEPRECATED
Deprecated in favor of channel_axis.
Deprecated since version 22.02.00.
Notes
Unsharp masking is an image sharpening technique. It is a linear image operation, and numerically stable, unlike deconvolution which is an ill-posed problem. Because of this stability, it is often preferred over deconvolution.
The main idea is as follows: sharp details are identified as the difference between the original image and its blurred version. These details are added back to the original image after a scaling step:
enhanced image = original + amount * (original - blurred)
When applying this filter to several color layers independently, color bleeding may occur. More visually pleasing result can be achieved by processing only the brightness/lightness/intensity channel in a suitable color space such as HSV, HSL, YUV, or YCbCr.
Unsharp masking is described in most introductory digital image processing books. This implementation is based on [1].
References
- 1
Maria Petrou, Costas Petrou “Image Processing: The Fundamentals”, (2010), ed ii., page 357, ISBN 13: 9781119994398 DOI:10.1002/9781119994398
- 2
Wikipedia. Unsharp masking https://en.wikipedia.org/wiki/Unsharp_masking
Examples
>>> import cupy as cp >>> array = cp.ones(shape=(5,5), dtype=cp.uint8)*100 >>> array[2,2] = 120 >>> array array([[100, 100, 100, 100, 100], [100, 100, 100, 100, 100], [100, 100, 120, 100, 100], [100, 100, 100, 100, 100], [100, 100, 100, 100, 100]], dtype=uint8) >>> cp.around(unsharp_mask(array, radius=0.5, amount=2),2) array([[0.39, 0.39, 0.39, 0.39, 0.39], [0.39, 0.39, 0.38, 0.39, 0.39], [0.39, 0.38, 0.53, 0.38, 0.39], [0.39, 0.39, 0.38, 0.39, 0.39], [0.39, 0.39, 0.39, 0.39, 0.39]])
>>> array = cp.ones(shape=(5,5), dtype=cp.int8)*100 >>> array[2,2] = 127 >>> cp.around(unsharp_mask(array, radius=0.5, amount=2),2) array([[0.79, 0.79, 0.79, 0.79, 0.79], [0.79, 0.78, 0.75, 0.78, 0.79], [0.79, 0.75, 1. , 0.75, 0.79], [0.79, 0.78, 0.75, 0.78, 0.79], [0.79, 0.79, 0.79, 0.79, 0.79]])
>>> cp.around(unsharp_mask(array, radius=0.5, amount=2, ... preserve_range=True), ... 2) array([[100. , 100. , 99.99, 100. , 100. ], [100. , 99.39, 95.48, 99.39, 100. ], [ 99.99, 95.48, 147.59, 95.48, 99.99], [100. , 99.39, 95.48, 99.39, 100. ], [100. , 100. , 99.99, 100. , 100. ]])
- cucim.skimage.filters.wiener(data, impulse_response=None, filter_params={}, K=0.25, predefined_filter=None)#
Minimum Mean Square Error (Wiener) inverse filter.
- Parameters
- data(M,N) ndarray
Input data.
- Kfloat or (M,N) ndarray
Ratio between power spectrum of noise and undegraded image.
- impulse_responsecallable f(r, c, **filter_params)
Impulse response of the filter. See LPIFilter2D.__init__.
- filter_paramsdict
Additional keyword parameters to the impulse_response function.
- Other Parameters
- predefined_filterLPIFilter2D
If you need to apply the same filter multiple times over different images, construct the LPIFilter2D and specify it here.
- cucim.skimage.filters.window(window_type, shape, warp_kwargs=None)#
Return an n-dimensional window of a given size and dimensionality.
- Parameters
- window_typestring, float, or tuple
The type of window to be created. Any window type supported by
scipy.signal.get_windowis allowed here. See notes below for a current list, or the SciPy documentation for the version of SciPy on your machine.- shapetuple of int or int
The shape of the window along each axis. If an integer is provided, a 1D window is generated.
- warp_kwargsdict
Keyword arguments passed to skimage.transform.warp (e.g.,
warp_kwargs={'order':3}to change interpolation method).
- Returns
- nd_windowndarray
A window of the specified
shape.dtypeisnp.double.
Notes
This function is based on
scipy.signal.get_windowand thus can access all of the window types available to that function (e.g.,"hann","boxcar"). Note that certain window types require parameters that have to be supplied with the window name as a tuple (e.g.,("tukey", 0.8)). If only a float is supplied, it is interpreted as the beta parameter of the Kaiser window.See https://docs.scipy.org/doc/scipy/reference/generated/scipy.signal.windows.get_window.html for more details.
Note that this function generates a double precision array of the specified
shapeand can thus generate very large arrays that consume a large amount of available memory.The approach taken here to create nD windows is to first calculate the Euclidean distance from the center of the intended nD window to each position in the array. That distance is used to sample, with interpolation, from a 1D window returned from
scipy.signal.get_window. The method of interpolation can be changed with theorderkeyword argument passed to skimage.transform.warp.Some coordinates in the output window will be outside of the original signal; these will be filled in with zeros.
Window types: - boxcar - triang - blackman - hamming - hann - bartlett - flattop - parzen - bohman - blackmanharris - nuttall - barthann - kaiser (needs beta) - gaussian (needs standard deviation) - general_gaussian (needs power, width) - slepian (needs width) - dpss (needs normalized half-bandwidth) - chebwin (needs attenuation) - exponential (needs decay scale) - tukey (needs taper fraction)
References
- 1
Two-dimensional window design, Wikipedia, https://en.wikipedia.org/wiki/Two_dimensional_window_design
Examples
Return a Hann window with shape (512, 512):
>>> from cucim.skimage.filters import window >>> w = window('hann', (512, 512))
Return a Kaiser window with beta parameter of 16 and shape (256, 256, 35):
>>> w = window(16, (256, 256, 35))
Return a Tukey window with an alpha parameter of 0.8 and shape (100, 300):
>>> w = window(('tukey', 0.8), (100, 300))
measure#
- cucim.skimage.measure.approximate_polygon(coords, tolerance)#
Approximate a polygonal chain with the specified tolerance.
It is based on the Douglas-Peucker algorithm.
Note that the approximated polygon is always within the convex hull of the original polygon.
- Parameters
- coords(N, 2) array
Coordinate array.
- tolerancefloat
Maximum distance from original points of polygon to approximated polygonal chain. If tolerance is 0, the original coordinate array is returned.
- Returns
- coords(M, 2) array
Approximated polygonal chain where M <= N.
References
- cucim.skimage.measure.block_reduce(image, block_size=2, func=<function sum>, cval=0, func_kwargs=None)#
Downsample image by applying function func to local blocks.
This function is useful for max and mean pooling, for example.
- Parameters
- imagendarray
N-dimensional input image.
- block_sizearray_like or int
Array containing down-sampling integer factor along each axis. Default block_size is 2.
- funccallable
Function object which is used to calculate the return value for each local block. This function must implement an
axisparameter. Primary functions arenumpy.sum,numpy.min,numpy.max,numpy.meanandnumpy.median. See also func_kwargs.- cvalfloat
Constant padding value if image is not perfectly divisible by the block size.
- func_kwargsdict
Keyword arguments passed to func. Notably useful for passing dtype argument to
np.mean. Takes dictionary of inputs, e.g.:func_kwargs={'dtype': np.float16}).
- Returns
- imagendarray
Down-sampled image with same number of dimensions as input image.
Examples
>>> import cupy as cp >>> from cucim.skimage.measure import block_reduce >>> image = cp.arange(3*3*4).reshape(3, 3, 4) >>> image array([[[ 0, 1, 2, 3], [ 4, 5, 6, 7], [ 8, 9, 10, 11]], [[12, 13, 14, 15], [16, 17, 18, 19], [20, 21, 22, 23]], [[24, 25, 26, 27], [28, 29, 30, 31], [32, 33, 34, 35]]]) >>> block_reduce(image, block_size=(3, 3, 1), func=cp.mean) array([[[16., 17., 18., 19.]]]) >>> image_max1 = block_reduce(image, block_size=(1, 3, 4), func=cp.max) >>> image_max1 array([[[11]], [[23]], [[35]]]) >>> image_max2 = block_reduce(image, block_size=(3, 1, 4), func=cp.max) >>> image_max2 array([[[27], [31], [35]]])
- cucim.skimage.measure.blur_effect(image, h_size=11, channel_axis=None, reduce_func=<built-in function max>)#
Compute a metric that indicates the strength of blur in an image (0 for no blur, 1 for maximal blur).
- Parameters
- imagendarray
RGB or grayscale nD image. The input image is converted to grayscale before computing the blur metric.
- h_sizeint, optional
Size of the re-blurring filter.
- channel_axisint or None, optional
If None, the image is assumed to be grayscale (single-channel). Otherwise, this parameter indicates which axis of the array corresponds to color channels.
- reduce_funccallable, optional
Function used to calculate the aggregation of blur metrics along all axes. If set to None, the entire list is returned, where the i-th element is the blur metric along the i-th axis. This function should be a host function that operates on standard python floats.
- Returns
- blurfloat (0 to 1) or list of floats
Blur metric: by default, the maximum of blur metrics along all axes.
Notes
h_size must keep the same value in order to compare results between images. Most of the time, the default size (11) is enough. This means that the metric can clearly discriminate blur up to an average 11x11 filter; if blur is higher, the metric still gives good results but its values tend towards an asymptote.
References
- 1
Frederique Crete, Thierry Dolmiere, Patricia Ladret, and Marina Nicolas “The blur effect: perception and estimation with a new no-reference perceptual blur metric” Proc. SPIE 6492, Human Vision and Electronic Imaging XII, 64920I (2007) https://hal.archives-ouvertes.fr/hal-00232709 DOI:10.1117/12.702790
- cucim.skimage.measure.centroid(image)#
Return the (weighted) centroid of an image.
- Parameters
- imagearray
The input image.
- Returns
- centertuple of float, length
image.ndim The centroid of the (nonzero) pixels in
image.
- centertuple of float, length
Examples
>>> import cupy as cp >>> from cucim.skimage.measure import centroid >>> image = cp.zeros((20, 20), dtype=np.float64) >>> image[13:17, 13:17] = 0.5 >>> image[10:12, 10:12] = 1 >>> centroid(image) array([13.16666667, 13.16666667])
- cucim.skimage.measure.inertia_tensor(image, mu=None, *, xp=<module 'cupy' from '/opt/conda/envs/rapids/lib/python3.9/site-packages/cupy/__init__.py'>)#
Compute the inertia tensor of the input image.
- Parameters
- imagearray
The input image.
- muarray, optional
The pre-computed central moments of
image. The inertia tensor computation requires the central moments of the image. If an application requires both the central moments and the inertia tensor (for example, skimage.measure.regionprops), then it is more efficient to pre-compute them and pass them to the inertia tensor call.
- Returns
- Tarray, shape
(image.ndim, image.ndim) The inertia tensor of the input image. \(T_{i, j}\) contains the covariance of image intensity along axes \(i\) and \(j\).
- Tarray, shape
References
- 1
https://en.wikipedia.org/wiki/Moment_of_inertia#Inertia_tensor
- 2
Bernd Jähne. Spatio-Temporal Image Processing: Theory and Scientific Applications. (Chapter 8: Tensor Methods) Springer, 1993.
- cucim.skimage.measure.inertia_tensor_eigvals(image, mu=None, T=None, *, xp=<module 'cupy' from '/opt/conda/envs/rapids/lib/python3.9/site-packages/cupy/__init__.py'>)#
Compute the eigenvalues of the inertia tensor of the image.
The inertia tensor measures covariance of the image intensity along the image axes. (See inertia_tensor.) The relative magnitude of the eigenvalues of the tensor is thus a measure of the elongation of a (bright) object in the image.
- Parameters
- imagearray
The input image.
- muarray, optional
The pre-computed central moments of
image.- Tarray, shape
(image.ndim, image.ndim) The pre-computed inertia tensor. If
Tis given,muandimageare ignored.
- Returns
- eigvalslist of float, length
image.ndim The eigenvalues of the inertia tensor of
image, in descending order.
- eigvalslist of float, length
Notes
Computing the eigenvalues requires the inertia tensor of the input image. This is much faster if the central moments (
mu) are provided, or, alternatively, one can provide the inertia tensor (T) directly.
- cucim.skimage.measure.label(label_image, background=None, return_num=False, connectivity=None)#
Label connected regions of an integer array.
Two pixels are connected when they are neighbors and have the same value. In 2D, they can be neighbors either in a 1- or 2-connected sense. The value refers to the maximum number of orthogonal hops to consider a pixel/voxel a neighbor:
1-connectivity 2-connectivity diagonal connection close-up [ ] [ ] [ ] [ ] [ ] | \ | / | <- hop 2 [ ]--[x]--[ ] [ ]--[x]--[ ] [x]--[ ] | / | \ hop 1 [ ] [ ] [ ] [ ]
- Parameters
- label_imagendarray of dtype int
Image to label.
- backgroundint, optional
Consider all pixels with this value as background pixels, and label them as 0. By default, 0-valued pixels are considered as background pixels.
- return_numbool, optional
Whether to return the number of assigned labels.
- connectivityint, optional
Maximum number of orthogonal hops to consider a pixel/voxel as a neighbor. Accepted values are ranging from 1 to input.ndim. If
None, a full connectivity ofinput.ndimis used.
- Returns
- labelsndarray of dtype int
Labeled array, where all connected regions are assigned the same integer value.
- numint, optional
Number of labels, which equals the maximum label index and is only returned if return_num is True.
- Other Parameters
- inputDEPRECATED
Deprecated in favor of label_image.
Deprecated since version 22.02.00.
See also
Notes
Currently the cucim implementation of this function always uses 32-bit integers for the label array. This is done for performance. In the future 64-bit integer support may also be added for better skimage compatibility.
References
- 1
Christophe Fiorio and Jens Gustedt, “Two linear time Union-Find strategies for image processing”, Theoretical Computer Science 154 (1996), pp. 165-181.
- 2
Kensheng Wu, Ekow Otoo and Arie Shoshani, “Optimizing connected component labeling algorithms”, Paper LBNL-56864, 2005, Lawrence Berkeley National Laboratory (University of California), http://repositories.cdlib.org/lbnl/LBNL-56864
Examples
>>> import cupy as cp >>> x = cp.eye(3).astype(int) >>> print(x) [[1 0 0] [0 1 0] [0 0 1]] >>> print(label(x, connectivity=1)) [[1 0 0] [0 2 0] [0 0 3]] >>> print(label(x, connectivity=2)) [[1 0 0] [0 1 0] [0 0 1]] >>> print(label(x, background=-1)) [[1 2 2] [2 1 2] [2 2 1]] >>> x = cp.asarray([[1, 0, 0], ... [1, 1, 5], ... [0, 0, 0]]) >>> print(label(x)) [[1 0 0] [1 1 2] [0 0 0]]
- cucim.skimage.measure.moments(image, order=3)#
Calculate all raw image moments up to a certain order.
- The following properties can be calculated from raw image moments:
Area as:
M[0, 0].Centroid as: {
M[1, 0] / M[0, 0],M[0, 1] / M[0, 0]}.
Note that raw moments are neither translation, scale nor rotation invariant.
- Parameters
- imagenD double or uint8 array
Rasterized shape as image.
- orderint, optional
Maximum order of moments. Default is 3.
- Returns
- m(
order + 1,order + 1) array Raw image moments.
- m(
References
- 1
Wilhelm Burger, Mark Burge. Principles of Digital Image Processing: Core Algorithms. Springer-Verlag, London, 2009.
- 2
B. Jähne. Digital Image Processing. Springer-Verlag, Berlin-Heidelberg, 6. edition, 2005.
- 3
T. H. Reiss. Recognizing Planar Objects Using Invariant Image Features, from Lecture notes in computer science, p. 676. Springer, Berlin, 1993.
- 4
Examples
>>> import cupy as cp >>> from cucim.skimage.measure import moments >>> image = cp.zeros((20, 20), dtype=cp.float64) >>> image[13:17, 13:17] = 1 >>> M = moments(image) >>> centroid = (M[1, 0] / M[0, 0], M[0, 1] / M[0, 0]) >>> centroid (array(14.5), array(14.5))
- cucim.skimage.measure.moments_central(image, center=None, order=3, **kwargs)#
Calculate all central image moments up to a certain order.
The center coordinates (cr, cc) can be calculated from the raw moments as: {
M[1, 0] / M[0, 0],M[0, 1] / M[0, 0]}.Note that central moments are translation invariant but not scale and rotation invariant.
- Parameters
- imagenD double or uint8 array
Rasterized shape as image.
- centertuple of float, optional
Coordinates of the image centroid. This will be computed if it is not provided.
- orderint, optional
The maximum order of moments computed.
- Returns
- mu(
order + 1,order + 1) array Central image moments.
- mu(
References
- 1
Wilhelm Burger, Mark Burge. Principles of Digital Image Processing: Core Algorithms. Springer-Verlag, London, 2009.
- 2
B. Jähne. Digital Image Processing. Springer-Verlag, Berlin-Heidelberg, 6. edition, 2005.
- 3
T. H. Reiss. Recognizing Planar Objects Using Invariant Image Features, from Lecture notes in computer science, p. 676. Springer, Berlin, 1993.
- 4
Examples
>>> import cupy as cp >>> from cucim.skimage.measure import moments, moments_central >>> image = cp.zeros((20, 20), dtype=cp.float64) >>> image[13:17, 13:17] = 1 >>> M = moments(image) >>> centroid = (M[1, 0] / M[0, 0], M[0, 1] / M[0, 0]) >>> moments_central(image, centroid) array([[16., 0., 20., 0.], [ 0., 0., 0., 0.], [20., 0., 25., 0.], [ 0., 0., 0., 0.]])
- cucim.skimage.measure.moments_coords(coords, order=3)#
Calculate all raw image moments up to a certain order.
- The following properties can be calculated from raw image moments:
Area as:
M[0, 0].Centroid as: {
M[1, 0] / M[0, 0],M[0, 1] / M[0, 0]}.
Note that raw moments are neither translation, scale nor rotation invariant.
- Parameters
- coords(N, D) double or uint8 array
Array of N points that describe an image of D dimensionality in Cartesian space.
- orderint, optional
Maximum order of moments. Default is 3.
- Returns
- M(
order + 1,order + 1, …) array Raw image moments. (D dimensions)
- M(
References
- 1
Johannes Kilian. Simple Image Analysis By Moments. Durham University, version 0.2, Durham, 2001.
Examples
>>> import cupy as cp >>> from cucim.skimage.measure import moments_coords >>> coords = cp.array([[row, col] ... for row in range(13, 17) ... for col in range(14, 18)], dtype=cp.float64) >>> M = moments_coords(coords) >>> centroid = (M[1, 0] / M[0, 0], M[0, 1] / M[0, 0]) >>> centroid (array(14.5), array(15.5))
- cucim.skimage.measure.moments_coords_central(coords, center=None, order=3)#
Calculate all central image moments up to a certain order.
- The following properties can be calculated from raw image moments:
Area as:
M[0, 0].Centroid as: {
M[1, 0] / M[0, 0],M[0, 1] / M[0, 0]}.
Note that raw moments are neither translation, scale nor rotation invariant.
- Parameters
- coords(N, D) double or uint8 array
Array of N points that describe an image of D dimensionality in Cartesian space. A tuple of coordinates as returned by
cp.nonzerois also accepted as input.- centertuple of float, optional
Coordinates of the image centroid. This will be computed if it is not provided.
- orderint, optional
Maximum order of moments. Default is 3.
- Returns
- Mc(
order + 1,order + 1, …) array Central image moments. (D dimensions)
- Mc(
References
- 1
Johannes Kilian. Simple Image Analysis By Moments. Durham University, version 0.2, Durham, 2001.
Examples
>>> import cupy as cp >>> from cucim.skimage.measure import moments_coords_central >>> coords = cp.array([[row, col] ... for row in range(13, 17) ... for col in range(14, 18)]) >>> moments_coords_central(coords) array([[16., 0., 20., 0.], [ 0., 0., 0., 0.], [20., 0., 25., 0.], [ 0., 0., 0., 0.]])
As seen above, for symmetric objects, odd-order moments (columns 1 and 3, rows 1 and 3) are zero when centered on the centroid, or center of mass, of the object (the default). If we break the symmetry by adding a new point, this no longer holds:
>>> coords2 = cp.concatenate((coords, cp.array([[17, 17]])), axis=0) >>> cp.round(moments_coords_central(coords2), ... decimals=2) array([[17. , 0. , 22.12, -2.49], [ 0. , 3.53, 1.73, 7.4 ], [25.88, 6.02, 36.63, 8.83], [ 4.15, 19.17, 14.8 , 39.6 ]])
Image moments and central image moments are equivalent (by definition) when the center is (0, 0):
>>> cp.allclose(moments_coords(coords), ... moments_coords_central(coords, (0, 0))) array(True)
- cucim.skimage.measure.moments_hu(nu)#
Calculate Hu’s set of image moments (2D-only).
Note that this set of moments is proofed to be translation, scale and rotation invariant.
- Parameters
- nu(M, M) array
Normalized central image moments, where M must be >= 4.
- Returns
- nu(7,) array
Hu’s set of image moments.
Notes
Due to the small array sizes, this function will be faster on the CPU. Consider transfering
nuto the host and runningskimage.measure.moments_huif the moments are not needed on the device.References
- 1
M. K. Hu, “Visual Pattern Recognition by Moment Invariants”, IRE Trans. Info. Theory, vol. IT-8, pp. 179-187, 1962
- 2
Wilhelm Burger, Mark Burge. Principles of Digital Image Processing: Core Algorithms. Springer-Verlag, London, 2009.
- 3
B. Jähne. Digital Image Processing. Springer-Verlag, Berlin-Heidelberg, 6. edition, 2005.
- 4
T. H. Reiss. Recognizing Planar Objects Using Invariant Image Features, from Lecture notes in computer science, p. 676. Springer, Berlin, 1993.
- 5
Examples
>>> import cupy as cp >>> from cucim.skimage.measure import (moments_central, moments_hu, ... moments_normalized) >>> image = cp.zeros((20, 20), dtype=np.float64) >>> image[13:17, 13:17] = 0.5 >>> image[10:12, 10:12] = 1 >>> mu = moments_central(image) >>> nu = moments_normalized(mu) >>> moments_hu(nu) array([7.45370370e-01, 3.51165981e-01, 1.04049179e-01, 4.06442107e-02, 2.64312299e-03, 2.40854582e-02, 6.50521303e-19])
- cucim.skimage.measure.moments_normalized(mu, order=3)#
Calculate all normalized central image moments up to a certain order.
Note that normalized central moments are translation and scale invariant but not rotation invariant.
- Parameters
- mu(M,[ …,] M) array
Central image moments, where M must be greater than or equal to
order.- orderint, optional
Maximum order of moments. Default is 3.
- Returns
- nu(
order + 1,[ …,]order + 1) array Normalized central image moments.
- nu(
Notes
Due to the small array sizes, this function should be faster on the CPU. Consider transfering
muto the host and runningskimage.measure.moments_normalized.References
- 1
Wilhelm Burger, Mark Burge. Principles of Digital Image Processing: Core Algorithms. Springer-Verlag, London, 2009.
- 2
B. Jähne. Digital Image Processing. Springer-Verlag, Berlin-Heidelberg, 6. edition, 2005.
- 3
T. H. Reiss. Recognizing Planar Objects Using Invariant Image Features, from Lecture notes in computer science, p. 676. Springer, Berlin, 1993.
- 4
Examples
>>> import cupy as cp >>> from cucim.skimage.measure import (moments, moments_central, ... moments_normalized) >>> image = cp.zeros((20, 20), dtype=cp.float64) >>> image[13:17, 13:17] = 1 >>> m = moments(image) >>> centroid = (m[0, 1] / m[0, 0], m[1, 0] / m[0, 0]) >>> mu = moments_central(image, centroid) >>> moments_normalized(mu) array([[ nan, nan, 0.078125 , 0. ], [ nan, 0. , 0. , 0. ], [0.078125 , 0. , 0.00610352, 0. ], [0. , 0. , 0. , 0. ]])
- cucim.skimage.measure.perimeter(image, neighbourhood=4)#
Calculate total perimeter of all objects in binary image.
- Parameters
- image(N, M) ndarray
2D binary image.
- neighbourhood4 or 8, optional
Neighborhood connectivity for border pixel determination. It is used to compute the contour. A higher neighbourhood widens the border on which the perimeter is computed.
- Returns
- perimeterfloat
Total perimeter of all objects in binary image.
References
- 1
K. Benkrid, D. Crookes. Design and FPGA Implementation of a Perimeter Estimator. The Queen’s University of Belfast. http://www.cs.qub.ac.uk/~d.crookes/webpubs/papers/perimeter.doc
Examples
>>> import cupy as cp >>> from skimage import data >>> from cucim.skimage import util >>> from cucim.skimage.measure import label >>> # coins image (binary) >>> img_coins = cp.array(data.coins() > 110) >>> # total perimeter of all objects in the image >>> perimeter(img_coins, neighbourhood=4) array(7796.86799644) >>> perimeter(img_coins, neighbourhood=8) array(8806.26807333)
- cucim.skimage.measure.profile_line(image, src, dst, linewidth=1, order=None, mode='reflect', cval=0.0, *, reduce_func=<function mean>)#
Return the intensity profile of an image measured along a scan line.
- Parameters
- imagendarray, shape (M, N[, C])
The image, either grayscale (2D array) or multichannel (3D array, where the final axis contains the channel information).
- srcarray_like, shape (2, )
The coordinates of the start point of the scan line.
- dstarray_like, shape (2, )
The coordinates of the end point of the scan line. The destination point is included in the profile, in contrast to standard numpy indexing.
- linewidthint, optional
Width of the scan, perpendicular to the line
- orderint in {0, 1, 2, 3, 4, 5}, optional
The order of the spline interpolation, default is 0 if image.dtype is bool and 1 otherwise. The order has to be in the range 0-5. See skimage.transform.warp for detail.
- mode{‘constant’, ‘nearest’, ‘reflect’, ‘mirror’, ‘wrap’}, optional
How to compute any values falling outside of the image.
- cvalfloat, optional
If mode is ‘constant’, what constant value to use outside the image.
- reduce_funccallable, optional
Function used to calculate the aggregation of pixel values perpendicular to the profile_line direction when linewidth > 1. If set to None the unreduced array will be returned.
- Returns
- return_valuearray
The intensity profile along the scan line. The length of the profile is the ceil of the computed length of the scan line.
Examples
>>> import cupy as cp >>> x = cp.asarray([[1, 1, 1, 2, 2, 2]]) >>> img = cp.vstack([cp.zeros_like(x), x, x, x, cp.zeros_like(x)]) >>> img array([[0, 0, 0, 0, 0, 0], [1, 1, 1, 2, 2, 2], [1, 1, 1, 2, 2, 2], [1, 1, 1, 2, 2, 2], [0, 0, 0, 0, 0, 0]]) >>> profile_line(img, (2, 1), (2, 4)) array([1., 1., 2., 2.]) >>> profile_line(img, (1, 0), (1, 6), cval=4) array([1., 1., 1., 2., 2., 2., 2.])
The destination point is included in the profile, in contrast to standard numpy indexing. For example:
>>> profile_line(img, (1, 0), (1, 6)) # The final point is out of bounds array([1., 1., 1., 2., 2., 2., 2.]) >>> profile_line(img, (1, 0), (1, 5)) # This accesses the full first row array([1., 1., 1., 2., 2., 2.])
For different reduce_func inputs:
>>> profile_line(img, (1, 0), (1, 3), linewidth=3, reduce_func=cp.mean) array([0.66666667, 0.66666667, 0.66666667, 1.33333333]) >>> profile_line(img, (1, 0), (1, 3), linewidth=3, reduce_func=cp.max) array([1, 1, 1, 2]) >>> profile_line(img, (1, 0), (1, 3), linewidth=3, reduce_func=cp.sum) array([2, 2, 2, 4])
The unreduced array will be returned when reduce_func is None or when reduce_func acts on each pixel value individually.
>>> profile_line(img, (1, 2), (4, 2), linewidth=3, order=0, ... reduce_func=None) array([[1, 1, 2], [1, 1, 2], [1, 1, 2], [0, 0, 0]]) >>> profile_line(img, (1, 0), (1, 3), linewidth=3, reduce_func=cp.sqrt) array([[1. , 1. , 0. ], [1. , 1. , 0. ], [1. , 1. , 0. ], [1.41421356, 1.41421356, 0. ]])
- cucim.skimage.measure.regionprops(label_image, intensity_image=None, cache=True, coordinates=None, *, extra_properties=None)#
Measure properties of labeled image regions.
- Parameters
- label_image(M, N[, P]) ndarray
Labeled input image. Labels with value 0 are ignored.
Changed in version 0.14.1: Previously,
label_imagewas processed bynumpy.squeezeand so any number of singleton dimensions was allowed. This resulted in inconsistent handling of images with singleton dimensions. To recover the old behaviour, useregionprops(np.squeeze(label_image), ...).- intensity_image(M, N[, P][, C]) ndarray, optional
Intensity (i.e., input) image with same size as labeled image, plus optionally an extra dimension for multichannel data. Currently, this extra channel dimension, if present, must be the last axis. Default is None.
Changed in version 0.18.0: The ability to provide an extra dimension for channels was added.
- cachebool, optional
Determine whether to cache calculated properties. The computation is much faster for cached properties, whereas the memory consumption increases.
- coordinatesDEPRECATED
This argument is deprecated and will be removed in a future version of scikit-image.
See Coordinate conventions for more details.
Deprecated since version 0.16.0: Use “rc” coordinates everywhere. It may be sufficient to call
numpy.transposeon your label image to get the same values as 0.15 and earlier. However, for some properties, the transformation will be less trivial. For example, the new orientation is \(\frac{\pi}{2}\) plus the old orientation.- extra_propertiesIterable of callables
Add extra property computation functions that are not included with skimage. The name of the property is derived from the function name, the dtype is inferred by calling the function on a small sample. If the name of an extra property clashes with the name of an existing property the extra property wil not be visible and a UserWarning is issued. A property computation function must take a region mask as its first argument. If the property requires an intensity image, it must accept the intensity image as the second argument.
- Returns
- propertieslist of RegionProperties
Each item describes one labeled region, and can be accessed using the attributes listed below.
See also
Notes
The following properties can be accessed as attributes or keys:
- areaint
Number of pixels of the region.
- area_bboxint
Number of pixels of bounding box.
- area_convexint
Number of pixels of convex hull image, which is the smallest convex polygon that encloses the region.
- area_filledint
Number of pixels of the region will all the holes filled in. Describes the area of the image_filled.
- axis_major_lengthfloat
The length of the major axis of the ellipse that has the same normalized second central moments as the region.
- axis_minor_lengthfloat
The length of the minor axis of the ellipse that has the same normalized second central moments as the region.
- bboxtuple
Bounding box
(min_row, min_col, max_row, max_col). Pixels belonging to the bounding box are in the half-open interval[min_row; max_row)and[min_col; max_col).- centroidarray
Centroid coordinate tuple
(row, col).- centroid_localarray
Centroid coordinate tuple
(row, col), relative to region bounding box.- centroid_weightedarray
Centroid coordinate tuple
(row, col)weighted with intensity image.- centroid_weighted_localarray
Centroid coordinate tuple
(row, col), relative to region bounding box, weighted with intensity image.- coords(N, 2) ndarray
Coordinate list
(row, col)of the region.- eccentricityfloat
Eccentricity of the ellipse that has the same second-moments as the region. The eccentricity is the ratio of the focal distance (distance between focal points) over the major axis length. The value is in the interval [0, 1). When it is 0, the ellipse becomes a circle.
- equivalent_diameter_areafloat
The diameter of a circle with the same area as the region.
- euler_numberint
Euler characteristic of the set of non-zero pixels. Computed as number of connected components subtracted by number of holes (input.ndim connectivity). In 3D, number of connected components plus number of holes subtracted by number of tunnels.
- extentfloat
Ratio of pixels in the region to pixels in the total bounding box. Computed as
area / (rows * cols)- feret_diameter_maxfloat
Maximum Feret’s diameter computed as the longest distance between points around a region’s convex hull contour as determined by
find_contours. [5]- image(H, J) ndarray
Sliced binary region image which has the same size as bounding box.
- image_convex(H, J) ndarray
Binary convex hull image which has the same size as bounding box.
- image_filled(H, J) ndarray
Binary region image with filled holes which has the same size as bounding box.
- image_intensityndarray
Image inside region bounding box.
- inertia_tensorndarray
Inertia tensor of the region for the rotation around its mass.
- inertia_tensor_eigvalstuple
The eigenvalues of the inertia tensor in decreasing order.
- intensity_maxfloat
Value with the greatest intensity in the region.
- intensity_meanfloat
Value with the mean intensity in the region.
- intensity_minfloat
Value with the least intensity in the region.
- labelint
The label in the labeled input image.
- moments(3, 3) ndarray
Spatial moments up to 3rd order:
m_ij = sum{ array(row, col) * row^i * col^j }
where the sum is over the row, col coordinates of the region.
- moments_central(3, 3) ndarray
Central moments (translation invariant) up to 3rd order:
mu_ij = sum{ array(row, col) * (row - row_c)^i * (col - col_c)^j }
where the sum is over the row, col coordinates of the region, and row_c and col_c are the coordinates of the region’s centroid.
- moments_hutuple
Hu moments (translation, scale and rotation invariant).
- moments_normalized(3, 3) ndarray
Normalized moments (translation and scale invariant) up to 3rd order:
nu_ij = mu_ij / m_00^[(i+j)/2 + 1]
where m_00 is the zeroth spatial moment.
- moments_weighted(3, 3) ndarray
Spatial moments of intensity image up to 3rd order:
wm_ij = sum{ array(row, col) * row^i * col^j }
where the sum is over the row, col coordinates of the region.
- moments_weighted_central(3, 3) ndarray
Central moments (translation invariant) of intensity image up to 3rd order:
wmu_ij = sum{ array(row, col) * (row - row_c)^i * (col - col_c)^j }
where the sum is over the row, col coordinates of the region, and row_c and col_c are the coordinates of the region’s weighted centroid.
- moments_weighted_hutuple
Hu moments (translation, scale and rotation invariant) of intensity image.
- moments_weighted_normalized(3, 3) ndarray
Normalized moments (translation and scale invariant) of intensity image up to 3rd order:
wnu_ij = wmu_ij / wm_00^[(i+j)/2 + 1]
where
wm_00is the zeroth spatial moment (intensity-weighted area).- orientationfloat
Angle between the 0th axis (rows) and the major axis of the ellipse that has the same second moments as the region, ranging from -pi/2 to pi/2 counter-clockwise.
- perimeterfloat
Perimeter of object which approximates the contour as a line through the centers of border pixels using a 4-connectivity.
- perimeter_croftonfloat
Perimeter of object approximated by the Crofton formula in 4 directions.
- slicetuple of slices
A slice to extract the object from the source image.
- solidityfloat
Ratio of pixels in the region to pixels of the convex hull image.
Each region also supports iteration, so that you can do:
for prop in region: print(prop, region[prop])
References
- 1
Wilhelm Burger, Mark Burge. Principles of Digital Image Processing: Core Algorithms. Springer-Verlag, London, 2009.
- 2
B. Jähne. Digital Image Processing. Springer-Verlag, Berlin-Heidelberg, 6. edition, 2005.
- 3
T. H. Reiss. Recognizing Planar Objects Using Invariant Image Features, from Lecture notes in computer science, p. 676. Springer, Berlin, 1993.
- 4
- 5
W. Pabst, E. Gregorová. Characterization of particles and particle systems, pp. 27-28. ICT Prague, 2007. https://old.vscht.cz/sil/keramika/Characterization_of_particles/CPPS%20_English%20version_.pdf
Examples
>>> from skimage import data, util >>> from cucim.skimage.measure import label, regionprops >>> img = cp.asarray(util.img_as_ubyte(data.coins()) > 110) >>> label_img = label(img, connectivity=img.ndim) >>> props = regionprops(label_img) >>> # centroid of first labeled object >>> props[0].centroid (22.72987986048314, 81.91228523446583) >>> # centroid of first labeled object >>> props[0]['centroid'] (22.72987986048314, 81.91228523446583)
Add custom measurements by passing functions as
extra_properties>>> from skimage import data, util >>> from cucim.skimage.measure import label, regionprops >>> import numpy as np >>> img = cp.asarray(util.img_as_ubyte(data.coins()) > 110) >>> label_img = label(img, connectivity=img.ndim) >>> def pixelcount(regionmask): ... return np.sum(regionmask) >>> props = regionprops(label_img, extra_properties=(pixelcount,)) >>> props[0].pixelcount array(7741) >>> props[1]['pixelcount'] array(42)
- cucim.skimage.measure.regionprops_table(label_image, intensity_image=None, properties=('label', 'bbox'), *, cache=True, separator='-', extra_properties=None)#
Compute image properties and return them as a pandas-compatible table.
The table is a dictionary mapping column names to value arrays. See Notes section below for details.
New in version 0.16.
- Parameters
- label_image(N, M[, P]) ndarray
Labeled input image. Labels with value 0 are ignored.
- intensity_image(M, N[, P][, C]) ndarray, optional
Intensity (i.e., input) image with same size as labeled image, plus optionally an extra dimension for multichannel data. Currently, this extra channel dimension, if present, must be the last axis. Default is None.
Changed in version 0.18.0: The ability to provide an extra dimension for channels was added.
- propertiestuple or list of str, optional
Properties that will be included in the resulting dictionary For a list of available properties, please see
regionprops(). Users should remember to add “label” to keep track of region identities.- cachebool, optional
Determine whether to cache calculated properties. The computation is much faster for cached properties, whereas the memory consumption increases.
- separatorstr, optional
For non-scalar properties not listed in OBJECT_COLUMNS, each element will appear in its own column, with the index of that element separated from the property name by this separator. For example, the inertia tensor of a 2D region will appear in four columns:
inertia_tensor-0-0,inertia_tensor-0-1,inertia_tensor-1-0, andinertia_tensor-1-1(where the separator is-).Object columns are those that cannot be split in this way because the number of columns would change depending on the object. For example,
imageandcoords.- extra_propertiesIterable of callables
Add extra property computation functions that are not included with skimage. The name of the property is derived from the function name, the dtype is inferred by calling the function on a small sample. If the name of an extra property clashes with the name of an existing property the extra property wil not be visible and a UserWarning is issued. A property computation function must take a region mask as its first argument. If the property requires an intensity image, it must accept the intensity image as the second argument.
- Returns
- out_dictdict
Dictionary mapping property names to an array of values of that property, one value per region. This dictionary can be used as input to pandas
DataFrameto map property names to columns in the frame and regions to rows. If the image has no regions, the arrays will have length 0, but the correct type.
Notes
Each column contains either a scalar property, an object property, or an element in a multidimensional array.
Properties with scalar values for each region, such as “eccentricity”, will appear as a float or int array with that property name as key.
Multidimensional properties of fixed size for a given image dimension, such as “centroid” (every centroid will have three elements in a 3D image, no matter the region size), will be split into that many columns, with the name {property_name}{separator}{element_num} (for 1D properties), {property_name}{separator}{elem_num0}{separator}{elem_num1} (for 2D properties), and so on.
For multidimensional properties that don’t have a fixed size, such as “image” (the image of a region varies in size depending on the region size), an object array will be used, with the corresponding property name as the key.
Examples
>>> from skimage import data, util, measure >>> image = data.coins() >>> label_image = measure.label(image > 110, connectivity=image.ndim) >>> props = measure.regionprops_table(label_image, image, ... properties=['label', 'inertia_tensor', ... 'inertia_tensor_eigvals']) >>> props {'label': array([ 1, 2, ...]), ... 'inertia_tensor-0-0': array([ 4.012...e+03, 8.51..., ...]), ... ..., 'inertia_tensor_eigvals-1': array([ 2.67...e+02, 2.83..., ...])}
The resulting dictionary can be directly passed to pandas, if installed, to obtain a clean DataFrame:
>>> import pandas as pd >>> data = pd.DataFrame(props) >>> data.head() label inertia_tensor-0-0 ... inertia_tensor_eigvals-1 0 1 4012.909888 ... 267.065503 1 2 8.514739 ... 2.834806 2 3 0.666667 ... 0.000000 3 4 0.000000 ... 0.000000 4 5 0.222222 ... 0.111111
[5 rows x 7 columns]
If we want to measure a feature that does not come as a built-in property, we can define custom functions and pass them as
extra_properties. For example, we can create a custom function that measures the intensity quartiles in a region:>>> from skimage import data, util, measure >>> import numpy as np >>> def quartiles(regionmask, intensity): ... return np.percentile(intensity[regionmask], q=(25, 50, 75)) >>> >>> image = data.coins() >>> label_image = measure.label(image > 110, connectivity=image.ndim) >>> props = measure.regionprops_table(label_image, intensity_image=image, ... properties=('label',), ... extra_properties=(quartiles,)) >>> import pandas as pd >>> pd.DataFrame(props).head() label quartiles-0 quartiles-1 quartiles-2 0 1 117.00 123.0 130.0 1 2 111.25 112.0 114.0 2 3 111.00 111.0 111.0 3 4 111.00 111.5 112.5 4 5 112.50 113.0 114.0
- cucim.skimage.measure.shannon_entropy(image, base=2)#
Calculate the Shannon entropy of an image.
The Shannon entropy is defined as S = -sum(pk * log(pk)), where pk are frequency/probability of pixels of value k.
- Parameters
- image(N, M) ndarray
Grayscale input image.
- basefloat, optional
The logarithmic base to use.
- Returns
- entropy0-dimensional float cupy.ndarray
Notes
The returned value is measured in bits or shannon (Sh) for base=2, natural unit (nat) for base=np.e and hartley (Hart) for base=10.
References
- 1
https://en.wikipedia.org/wiki/Entropy_(information_theory) <https://en.wikipedia.org/wiki/Entropy_(information_theory)>`_ # noqa
- 2
Examples
>>> import cupy as cp >>> from skimage import data >>> from cucim.skimage.measure import shannon_entropy >>> shannon_entropy(cp.array(data.camera())) array(7.23169501)
- cucim.skimage.measure.subdivide_polygon(coords, degree=2, preserve_ends=False)#
Subdivision of polygonal curves using B-Splines.
Note that the resulting curve is always within the convex hull of the original polygon. Circular polygons stay closed after subdivision.
- Parameters
- coords(N, 2) array
Coordinate array.
- degree{1, 2, 3, 4, 5, 6, 7}, optional
Degree of B-Spline. Default is 2.
- preserve_endsbool, optional
Preserve first and last coordinate of non-circular polygon. Default is False.
- Returns
- coords(M, 2) array
Subdivided coordinate array.
References
metrics#
- cucim.skimage.metrics.mean_squared_error(image0, image1)#
Compute the mean-squared error between two images.
- Parameters
- image0, image1ndarray
Images. Any dimensionality, must have same shape.
- Returns
- msefloat
The mean-squared error (MSE) metric.
Notes
Changed in version 0.16: This function was renamed from
skimage.measure.compare_msetoskimage.metrics.mean_squared_error.
- cucim.skimage.metrics.normalized_mutual_information(image0, image1, *, bins=100)#
Compute the normalized mutual information (NMI).
The normalized mutual information of \(A\) and \(B\) is given by:
..math::
Y(A, B) = frac{H(A) + H(B)}{H(A, B)}
where \(H(X) := - \sum_{x \in X}{x \log x}\) is the entropy.
It was proposed to be useful in registering images by Colin Studholme and colleagues [1]. It ranges from 1 (perfectly uncorrelated image values) to 2 (perfectly correlated image values, whether positively or negatively).
- Parameters
- image0, image1ndarray
Images to be compared. The two input images must have the same number of dimensions.
- binsint or sequence of int, optional
The number of bins along each axis of the joint histogram.
- Returns
- nmifloat
The normalized mutual information between the two arrays, computed at the granularity given by
bins. Higher NMI implies more similar input images.
- Raises
- ValueError
If the images don’t have the same number of dimensions.
Notes
If the two input images are not the same shape, the smaller image is padded with zeros.
References
- 1
C. Studholme, D.L.G. Hill, & D.J. Hawkes (1999). An overlap invariant entropy measure of 3D medical image alignment. Pattern Recognition 32(1):71-86 DOI:10.1016/S0031-3203(98)00091-0
- cucim.skimage.metrics.normalized_root_mse(image_true, image_test, *, normalization='euclidean')#
Compute the normalized root mean-squared error (NRMSE) between two images.
- Parameters
- image_truendarray
Ground-truth image, same shape as im_test.
- image_testndarray
Test image.
- normalization{‘euclidean’, ‘min-max’, ‘mean’}, optional
Controls the normalization method to use in the denominator of the NRMSE. There is no standard method of normalization across the literature [1]. The methods available here are as follows:
‘euclidean’ : normalize by the averaged Euclidean norm of
im_true:NRMSE = RMSE * sqrt(N) / || im_true ||
where || . || denotes the Frobenius norm and
N = im_true.size. This result is equivalent to:NRMSE = || im_true - im_test || / || im_true ||.
‘min-max’ : normalize by the intensity range of
im_true.‘mean’ : normalize by the mean of
im_true
- Returns
- nrmsefloat
The NRMSE metric.
Notes
Changed in version 0.16: This function was renamed from
skimage.measure.compare_nrmsetoskimage.metrics.normalized_root_mse.References
- cucim.skimage.metrics.peak_signal_noise_ratio(image_true, image_test, *, data_range=None)#
Compute the peak signal to noise ratio (PSNR) for an image.
- Parameters
- image_truendarray
Ground-truth image, same shape as im_test.
- image_testndarray
Test image.
- data_rangeint, optional
The data range of the input image (distance between minimum and maximum possible values). By default, this is estimated from the image data-type.
- Returns
- psnrfloat
The PSNR metric.
Notes
Changed in version 0.16: This function was renamed from
skimage.measure.compare_psnrtoskimage.metrics.peak_signal_noise_ratio.References
- cucim.skimage.metrics.structural_similarity(im1, im2, *, win_size=None, gradient=False, data_range=None, channel_axis=None, multichannel=False, gaussian_weights=False, full=False, **kwargs)#
Compute the mean structural similarity index between two images.
- Parameters
- im1, im2ndarray
Images. Any dimensionality with same shape.
- win_sizeint or None, optional
The side-length of the sliding window used in comparison. Must be an odd value. If gaussian_weights is True, this is ignored and the window size will depend on sigma.
- gradientbool, optional
If True, also return the gradient with respect to im2.
- data_rangefloat, optional
The data range of the input image (distance between minimum and maximum possible values). By default, this is estimated from the image data-type.
- channel_axisint or None, optional
If None, the image is assumed to be a grayscale (single channel) image. Otherwise, this parameter indicates which axis of the array corresponds to channels.
- multichannelbool, optional
If True, treat the last dimension of the array as channels. Similarity calculations are done independently for each channel then averaged. This argument is deprecated: specify channel_axis instead.
- gaussian_weightsbool, optional
If True, each patch has its mean and variance spatially weighted by a normalized Gaussian kernel of width sigma=1.5.
- fullbool, optional
If True, also return the full structural similarity image.
- Returns
- mssimfloat
The mean structural similarity index over the image.
- gradndarray
The gradient of the structural similarity between im1 and im2 [2]. This is only returned if gradient is set to True.
- Sndarray
The full SSIM image. This is only returned if full is set to True.
- Other Parameters
- use_sample_covariancebool
If True, normalize covariances by N-1 rather than, N where N is the number of pixels within the sliding window.
- K1float
Algorithm parameter, K1 (small constant, see [1]).
- K2float
Algorithm parameter, K2 (small constant, see [1]).
- sigmafloat
Standard deviation for the Gaussian when gaussian_weights is True.
- multichannelDEPRECATED
Deprecated in favor of channel_axis.
Deprecated since version 22.02.00.
Notes
To match the implementation of Wang et. al. [1], set gaussian_weights to True, sigma to 1.5, and use_sample_covariance to False.
Changed in version 0.16: This function was renamed from
skimage.measure.compare_ssimtoskimage.metrics.structural_similarity.References
- 1(1,2,3)
Wang, Z., Bovik, A. C., Sheikh, H. R., & Simoncelli, E. P. (2004). Image quality assessment: From error visibility to structural similarity. IEEE Transactions on Image Processing, 13, 600-612. https://ece.uwaterloo.ca/~z70wang/publications/ssim.pdf, DOI:10.1109/TIP.2003.819861
- 2
Avanaki, A. N. (2009). Exact global histogram specification optimized for structural similarity. Optical Review, 16, 613-621. arXiv:0901.0065 DOI:10.1007/s10043-009-0119-z
morphology#
- cucim.skimage.morphology.ball(radius, dtype=<class 'numpy.uint8'>, *, strict_radius=True, decomposition=None)#
Generates a ball-shaped footprint.
This is the 3D equivalent of a disk. A pixel is within the neighborhood if the Euclidean distance between it and the origin is no greater than radius.
- Parameters
- radiusint
The radius of the ball-shaped footprint.
- Returns
- footprintcupy.ndarray
The footprint where elements of the neighborhood are 1 and 0 otherwise.
- Other Parameters
- dtypedata-type, optional
The data type of the footprint.
- strict_radiusbool, optional
If False, extend the radius by 0.5. This allows the circle to expand further within a cube that remains of size
2 * radius + 1along each axis. This parameter is ignored if decomposition is not None.- decomposition{None, ‘sequence’}, optional
If None, a single array is returned. For ‘sequence’, a tuple of smaller footprints is returned. Applying this series of smaller footprints will given a result equivalent to a single, larger footprint, but with better computational performance. For ball footprints, the sequence decomposition is not exactly equivalent to decomposition=None. See Notes for more details.
Notes
The disk produced by the decomposition=’sequence’ mode is not identical to that with decomposition=None. Here we extend the approach taken in [1] for disks to the 3D case, using 3-dimensional extensions of the “square”, “diamond” and “t-shaped” elements from that publication. All of these elementary elements have size
(3,) * ndim. We numerically computed the number of repetitions of each element that gives the closest match to the ball computed with kwargsstrict_radius=False, decomposition=None.Empirically, the equivalent composite footprint to the sequence decomposition approaches a rhombicuboctahedron (26-faces [2]).
References
- 1
Park, H and Chin R.T. Decomposition of structuring elements for optimal implementation of morphological operations. In Proceedings: 1997 IEEE Workshop on Nonlinear Signal and Image Processing, London, UK. https://www.iwaenc.org/proceedings/1997/nsip97/pdf/scan/ns970226.pdf
- 2
- cucim.skimage.morphology.binary_closing(image, footprint=None, out=None)#
Return fast binary morphological closing of an image.
This function returns the same result as grayscale closing but performs faster for binary images.
The morphological closing on an image is defined as a dilation followed by an erosion. Closing can remove small dark spots (i.e. “pepper”) and connect small bright cracks. This tends to “close” up (dark) gaps between (bright) features.
- Parameters
- imagendarray
Binary input image.
- footprintndarray or tuple, optional
The neighborhood expressed as a 2-D array of 1’s and 0’s. If None, use a cross-shaped footprint (connectivity=1). The footprint can also be provided as a sequence of smaller footprints as described in the notes below.
- outndarray of bool, optional
The array to store the result of the morphology. If None, is passed, a new array will be allocated.
- Returns
- closingndarray of bool
The result of the morphological closing.
- Other Parameters
- selemDEPRECATED
Deprecated in favor of footprint.
Deprecated since version 22.02.00.
Notes
The footprint can also be a provided as a sequence of 2-tuples where the first element of each 2-tuple is a footprint ndarray and the second element is an integer describing the number of times it should be iterated. For example
footprint=[(cp.ones((9, 1)), 1), (cp.ones((1, 9)), 1)]would apply a 9x1 footprint followed by a 1x9 footprint resulting in a net effect that is the same asfootprint=cp.ones((9, 9)), but with lower computational cost. Most of the builtin footprints such asskimage.morphology.diskprovide an option to automically generate a footprint sequence of this type.
- cucim.skimage.morphology.binary_dilation(image, footprint=None, out=None)#
Return fast binary morphological dilation of an image.
This function returns the same result as grayscale dilation but performs faster for binary images.
Morphological dilation sets a pixel at
(i,j)to the maximum over all pixels in the neighborhood centered at(i,j). Dilation enlarges bright regions and shrinks dark regions.- Parameters
- imagendarray
Binary input image.
- footprintndarray or tuple, optional
The neighborhood expressed as a 2-D array of 1’s and 0’s. If None, use a cross-shaped footprint (connectivity=1). The footprint can also be provided as a sequence of smaller footprints as described in the notes below.
- outndarray of bool, optional
The array to store the result of the morphology. If None is passed, a new array will be allocated.
- Returns
- dilatedndarray of bool or uint
The result of the morphological dilation with values in
[False, True].
- Other Parameters
- selemDEPRECATED
Deprecated in favor of footprint.
Deprecated since version 22.02.00.
Notes
The footprint can also be a provided as a sequence of 2-tuples where the first element of each 2-tuple is a footprint ndarray and the second element is an integer describing the number of times it should be iterated. For example
footprint=[(cp.ones((9, 1)), 1), (cp.ones((1, 9)), 1)]would apply a 9x1 footprint followed by a 1x9 footprint resulting in a net effect that is the same asfootprint=cp.ones((9, 9)), but with lower computational cost. Most of the builtin footprints such asskimage.morphology.diskprovide an option to automically generate a footprint sequence of this type.
- cucim.skimage.morphology.binary_erosion(image, footprint=None, out=None)#
Return fast binary morphological erosion of an image.
This function returns the same result as grayscale erosion but performs faster for binary images.
Morphological erosion sets a pixel at
(i,j)to the minimum over all pixels in the neighborhood centered at(i,j). Erosion shrinks bright regions and enlarges dark regions.- Parameters
- imagendarray
Binary input image.
- footprintndarray or tuple, optional
The neighborhood expressed as a 2-D array of 1’s and 0’s. If None, use a cross-shaped footprint (connectivity=1). The footprint can also be provided as a sequence of smaller footprints as described in the notes below.
- outndarray of bool, optional
The array to store the result of the morphology. If None is passed, a new array will be allocated.
- Returns
- erodedndarray of bool or uint
The result of the morphological erosion taking values in
[False, True].
- Other Parameters
- selemDEPRECATED
Deprecated in favor of footprint.
Deprecated since version 22.02.00.
Notes
The footprint can also be a provided as a sequence of 2-tuples where the first element of each 2-tuple is a footprint ndarray and the second element is an integer describing the number of times it should be iterated. For example
footprint=[(cp.ones((9, 1)), 1), (cp.ones((1, 9)), 1)]would apply a 9x1 footprint followed by a 1x9 footprint resulting in a net effect that is the same asfootprint=cp.ones((9, 9)), but with lower computational cost. Most of the builtin footprints such asskimage.morphology.diskprovide an option to automically generate a footprint sequence of this type.
- cucim.skimage.morphology.binary_opening(image, footprint=None, out=None)#
Return fast binary morphological opening of an image.
This function returns the same result as grayscale opening but performs faster for binary images.
The morphological opening on an image is defined as an erosion followed by a dilation. Opening can remove small bright spots (i.e. “salt”) and connect small dark cracks. This tends to “open” up (dark) gaps between (bright) features.
- Parameters
- imagendarray
Binary input image.
- footprintndarray or tuple, optional
The neighborhood expressed as a 2-D array of 1’s and 0’s. If None, use a cross-shaped footprint (connectivity=1). The footprint can also be provided as a sequence of smaller footprints as described in the notes below.
- outndarray of bool, optional
The array to store the result of the morphology. If None is passed, a new array will be allocated.
- Returns
- openingndarray of bool
The result of the morphological opening.
- Other Parameters
- selemDEPRECATED
Deprecated in favor of footprint.
Deprecated since version 22.02.00.
Notes
The footprint can also be a provided as a sequence of 2-tuples where the first element of each 2-tuple is a footprint ndarray and the second element is an integer describing the number of times it should be iterated. For example
footprint=[(cp.ones((9, 1)), 1), (cp.ones((1, 9)), 1)]would apply a 9x1 footprint followed by a 1x9 footprint resulting in a net effect that is the same asfootprint=cp.ones((9, 9)), but with lower computational cost. Most of the builtin footprints such asskimage.morphology.diskprovide an option to automically generate a footprint sequence of this type.
- cucim.skimage.morphology.black_tophat(image, footprint=None, out=None)#
Return black top hat of an image.
The black top hat of an image is defined as its morphological closing minus the original image. This operation returns the dark spots of the image that are smaller than the footprint. Note that dark spots in the original image are bright spots after the black top hat.
- Parameters
- imagecupy.ndarray
Image array.
- footprintcupy.ndarray, optional
The neighborhood expressed as a 2-D array of 1’s and 0’s. If None, use a cross-shaped footprint (connectivity=1). The footprint can also be provided as a sequence of smaller footprints as described in the notes below.
- outcupy.ndarray, optional
The array to store the result of the morphology. If None is passed, a new array will be allocated.
- Returns
- outcupy.ndarray, same shape and type as image
The result of the morphological black top hat.
- Other Parameters
- selemDEPRECATED
Deprecated in favor of footprint.
Deprecated since version 22.02.00.
See also
Notes
The footprint can also be a provided as a sequence of 2-tuples where the first element of each 2-tuple is a footprint ndarray and the second element is an integer describing the number of times it should be iterated. For example
footprint=[(cp.ones((9, 1)), 1), (cp.ones((1, 9)), 1)]would apply a 9x1 footprint followed by a 1x9 footprint resulting in a net effect that is the same asfootprint=cp.ones((9, 9)), but with lower computational cost. Most of the builtin footprints such asskimage.morphology.diskprovide an option to automically generate a footprint sequence of this type.References
Examples
>>> # Change dark peak to bright peak and subtract background >>> import cupy as cp >>> from cucim.skimage.morphology import square >>> dark_on_grey = cp.asarray([[7, 6, 6, 6, 7], ... [6, 5, 4, 5, 6], ... [6, 4, 0, 4, 6], ... [6, 5, 4, 5, 6], ... [7, 6, 6, 6, 7]], dtype=cp.uint8) >>> black_tophat(dark_on_grey, square(3)) array([[0, 0, 0, 0, 0], [0, 0, 1, 0, 0], [0, 1, 5, 1, 0], [0, 0, 1, 0, 0], [0, 0, 0, 0, 0]], dtype=uint8)
- cucim.skimage.morphology.closing(image, footprint=None, out=None)#
Return grayscale morphological closing of an image.
The morphological closing of an image is defined as a dilation followed by an erosion. Closing can remove small dark spots (i.e. “pepper”) and connect small bright cracks. This tends to “close” up (dark) gaps between (bright) features.
- Parameters
- imagecupy.ndarray
Image array.
- footprintcupy.ndarray, optional
The neighborhood expressed as a 2-D array of 1’s and 0’s. If None, use a cross-shaped footprint (connectivity=1). The footprint can also be provided as a sequence of smaller footprints as described in the notes below.
- outcupy.ndarray, optional
The array to store the result of the morphology. If None, a new array will be allocated.
- Returns
- closingcupy.ndarray, same shape and type as image
The result of the morphological closing.
- Other Parameters
- selemDEPRECATED
Deprecated in favor of footprint.
Deprecated since version 22.02.00.
Notes
The footprint can also be a provided as a sequence of 2-tuples where the first element of each 2-tuple is a footprint ndarray and the second element is an integer describing the number of times it should be iterated. For example
footprint=[(cp.ones((9, 1)), 1), (cp.ones((1, 9)), 1)]would apply a 9x1 footprint followed by a 1x9 footprint resulting in a net effect that is the same asfootprint=cp.ones((9, 9)), but with lower computational cost. Most of the builtin footprints such asskimage.morphology.diskprovide an option to automically generate a footprint sequence of this type.Examples
>>> # Close a gap between two bright lines >>> import cupy as cp >>> from cucim.skimage.morphology import square >>> broken_line = cp.asarray([[0, 0, 0, 0, 0], ... [0, 0, 0, 0, 0], ... [1, 1, 0, 1, 1], ... [0, 0, 0, 0, 0], ... [0, 0, 0, 0, 0]], dtype=cp.uint8) >>> closing(broken_line, square(3)) array([[0, 0, 0, 0, 0], [0, 0, 0, 0, 0], [1, 1, 1, 1, 1], [0, 0, 0, 0, 0], [0, 0, 0, 0, 0]], dtype=uint8)
- cucim.skimage.morphology.cube(width, dtype=<class 'numpy.uint8'>, *, decomposition=None)#
Generates a cube-shaped footprint.
This is the 3D equivalent of a square. Every pixel along the perimeter has a chessboard distance no greater than radius (radius=floor(width/2)) pixels.
- Parameters
- widthint
The width, height and depth of the cube.
- Returns
- footprintcupy.ndarray
The footprint where elements of the neighborhood are 1 and 0 otherwise. When decomposition is None, this is just a numpy.ndarray. Otherwise, this will be a tuple whose length is equal to the number of unique structuring elements to apply (see Notes for more detail)
- Other Parameters
- dtypedata-type, optional
The data type of the footprint.
- decomposition{None, ‘separable’, ‘sequence’}, optional
If None, a single array is returned. For ‘sequence’, a tuple of smaller footprints is returned. Applying this series of smaller footprints will given an identical result to a single, larger footprint, but often with better computational performance. See Notes for more details.
Notes
When decomposition is not None, each element of the footprint tuple is a 2-tuple of the form
(ndarray, num_iter)that specifies a footprint array and the number of iterations it is to be applied.For binary morphology, using
decomposition='sequence'was observed to give better performance, with the magnitude of the performance increase rapidly increasing with footprint size. For grayscale morphology with square footprints, it is recommended to usedecomposition=Nonesince the internal SciPy functions that are called already have a fast implementation based on separable 1D sliding windows.The ‘sequence’ decomposition mode only supports odd valued width. If width is even, the sequence used will be identical to the ‘separable’ mode.
- cucim.skimage.morphology.diamond(radius, dtype=<class 'numpy.uint8'>, *, decomposition=None)#
Generates a flat, diamond-shaped footprint.
A pixel is part of the neighborhood (i.e. labeled 1) if the city block/Manhattan distance between it and the center of the neighborhood is no greater than radius.
- Parameters
- radiusint
The radius of the diamond-shaped footprint.
- Returns
- footprintcupy.ndarray
The footprint where elements of the neighborhood are 1 and 0 otherwise. When decomposition is None, this is just a numpy.ndarray. Otherwise, this will be a tuple whose length is equal to the number of unique structuring elements to apply (see Notes for more detail)
- Other Parameters
- dtypedata-type, optional
The data type of the footprint.
- decomposition{None, ‘sequence’}, optional
If None, a single array is returned. For ‘sequence’, a tuple of smaller footprints is returned. Applying this series of smaller footprints will given an identical result to a single, larger footprint, but with better computational performance. See Notes for more details.
Notes
When decomposition is not None, each element of the footprint tuple is a 2-tuple of the form
(ndarray, num_iter)that specifies a footprint array and the number of iterations it is to be applied.For either binary or grayscale morphology, using
decomposition='sequence'was observed to have a performance benefit, with the magnitude of the benefit increasing with increasing footprint size.
- cucim.skimage.morphology.dilation(image, footprint=None, out=None, shift_x=False, shift_y=False)#
Return grayscale morphological dilation of an image.
Morphological dilation sets the value of a pixel to the maximum over all pixel values within a local neighborhood centered about it. The values where the footprint is 1 define this neighborhood. Dilation enlarges bright regions and shrinks dark regions.
- Parameters
- imagecupy.ndarray
Image array.
- footprintcupy.ndarray, optional
The neighborhood expressed as a 2-D array of 1’s and 0’s. If None, use a cross-shaped footprint (connectivity=1). The footprint can also be provided as a sequence of smaller footprints as described in the notes below.
- outcupy.ndarray, optional
The array to store the result of the morphology. If None is passed, a new array will be allocated.
- shift_x, shift_ybool, optional
Shift footprint about center point. This only affects 2D eccentric footprints (i.e., footprints with even-numbered sides).
- Returns
- dilatedcupy.ndarray, same shape and type as image
The result of the morphological dilation.
- Other Parameters
- selemDEPRECATED
Deprecated in favor of footprint.
Deprecated since version 22.02.00.
Notes
For uint8 (and uint16 up to a certain bit-depth) data, the lower algorithm complexity makes the skimage.filters.rank.maximum function more efficient for larger images and footprints.
The footprint can also be a provided as a sequence of 2-tuples where the first element of each 2-tuple is a footprint ndarray and the second element is an integer describing the number of times it should be iterated. For example
footprint=[(cp.ones((9, 1)), 1), (cp.ones((1, 9)), 1)]would apply a 9x1 footprint followed by a 1x9 footprint resulting in a net effect that is the same asfootprint=cp.ones((9, 9)), but with lower computational cost. Most of the builtin footprints such asskimage.morphology.diskprovide an option to automically generate a footprint sequence of this type.Examples
>>> # Dilation enlarges bright regions >>> import cupy as cp >>> from cucim.skimage.morphology import square >>> bright_pixel = cp.asarray([[0, 0, 0, 0, 0], ... [0, 0, 0, 0, 0], ... [0, 0, 1, 0, 0], ... [0, 0, 0, 0, 0], ... [0, 0, 0, 0, 0]], dtype=cp.uint8) >>> dilation(bright_pixel, square(3)) array([[0, 0, 0, 0, 0], [0, 1, 1, 1, 0], [0, 1, 1, 1, 0], [0, 1, 1, 1, 0], [0, 0, 0, 0, 0]], dtype=uint8)
- cucim.skimage.morphology.disk(radius, dtype=<class 'numpy.uint8'>, *, strict_radius=True, decomposition=None)#
Generates a flat, disk-shaped footprint.
A pixel is within the neighborhood if the Euclidean distance between it and the origin is no greater than radius (This is only approximately True, when decomposition == ‘sequence’).
- Parameters
- radiusint
The radius of the disk-shaped footprint.
- Returns
- footprintcupy.ndarray
The footprint where elements of the neighborhood are 1 and 0 otherwise.
- Other Parameters
- dtypedata-type, optional
The data type of the footprint.
- strict_radiusbool, optional
If False, extend the radius by 0.5. This allows the circle to expand further within a cube that remains of size
2 * radius + 1along each axis. This parameter is ignored if decomposition is not None.- decomposition{None, ‘sequence’, ‘crosses’}, optional
If None, a single array is returned. For ‘sequence’, a tuple of smaller footprints is returned. Applying this series of smaller footprints will given a result equivalent to a single, larger footprint, but with better computational performance. For disk footprints, the ‘sequence’ or ‘crosses’ decompositions are not always exactly equivalent to
decomposition=None. See Notes for more details.
Notes
When decomposition is not None, each element of the footprint tuple is a 2-tuple of the form
(ndarray, num_iter)that specifies a footprint array and the number of iterations it is to be applied.The disk produced by the
decomposition='sequence'mode may not be identical to that withdecomposition=None. A disk footprint can be approximated by applying a series of smaller footprints of extent 3 along each axis. Specific solutions for this are given in [1] for the case of 2D disks with radius 2 through 10. Here, we numerically computed the number of repetitions of each element that gives the closest match to the disk computed with kwargsstrict_radius=False, decomposition=None.Empirically, the series decomposition at large radius approaches a hexadecagon (a 16-sided polygon [2]). In [3], the authors demonstrate that a hexadecagon is the closest approximation to a disk that can be achieved for decomposition with footprints of shape (3, 3).
The disk produced by the
decomposition='crosses'is often but not always identical to that withdecomposition=None. It tends to give a closer approximation thandecomposition='sequence', at a performance that is fairly comparable. The individual cross-shaped elements are not limited to extent (3, 3) in size. Unlike the ‘seqeuence’ decomposition, the ‘crosses’ decomposition can also accurately approximate the shape of disks withstrict_radius=True. The method is based on an adaption of algorithm 1 given in [4].References
- 1
Park, H and Chin R.T. Decomposition of structuring elements for optimal implementation of morphological operations. In Proceedings: 1997 IEEE Workshop on Nonlinear Signal and Image Processing, London, UK. https://www.iwaenc.org/proceedings/1997/nsip97/pdf/scan/ns970226.pdf
- 2
- 3
Vanrell, M and Vitrià, J. Optimal 3 × 3 decomposable disks for morphological transformations. Image and Vision Computing, Vol. 15, Issue 11, 1997. DOI:10.1016/S0262-8856(97)00026-7
- 4
Li, D. and Ritter, G.X. Decomposition of Separable and Symmetric Convex Templates. Proc. SPIE 1350, Image Algebra and Morphological Image Processing, (1 November 1990). DOI:10.1117/12.23608
- cucim.skimage.morphology.erosion(image, footprint=None, out=None, shift_x=False, shift_y=False)#
Return grayscale morphological erosion of an image.
Morphological erosion sets a pixel at (i,j) to the minimum over all pixels in the neighborhood centered at (i,j). Erosion shrinks bright regions and enlarges dark regions.
- Parameters
- imagecupy.ndarray
Image array.
- footprintcupy.ndarray, optional
The neighborhood expressed as a 2-D array of 1’s and 0’s. If None, use a cross-shaped footprint (connectivity=1). The footprint can also be provided as a sequence of smaller footprints as described in the notes below.
- outcupy.ndarray, optional
The array to store the result of the morphology. If None is passed, a new array will be allocated.
- shift_x, shift_ybool, optional
shift footprint about center point. This only affects eccentric footprints (i.e. footprint with even numbered sides).
- Returns
- erodedcupy.ndarray, same shape as image
The result of the morphological erosion.
- Other Parameters
- selemDEPRECATED
Deprecated in favor of footprint.
Deprecated since version 22.02.00.
Notes
For
uint8(anduint16up to a certain bit-depth) data, the lower algorithm complexity makes the skimage.filters.rank.minimum function more efficient for larger images and footprints.The footprint can also be a provided as a sequence of 2-tuples where the first element of each 2-tuple is a footprint ndarray and the second element is an integer describing the number of times it should be iterated. For example
footprint=[(cp.ones((9, 1)), 1), (cp.ones((1, 9)), 1)]would apply a 9x1 footprint followed by a 1x9 footprint resulting in a net effect that is the same asfootprint=cp.ones((9, 9)), but with lower computational cost. Most of the builtin footprints such asskimage.morphology.diskprovide an option to automically generate a footprint sequence of this type.Examples
>>> # Erosion shrinks bright regions >>> import cupy as cp >>> from cucim.skimage.morphology import square >>> bright_square = cp.asarray([[0, 0, 0, 0, 0], ... [0, 1, 1, 1, 0], ... [0, 1, 1, 1, 0], ... [0, 1, 1, 1, 0], ... [0, 0, 0, 0, 0]], dtype=cp.uint8) >>> erosion(bright_square, square(3)) array([[0, 0, 0, 0, 0], [0, 0, 0, 0, 0], [0, 0, 1, 0, 0], [0, 0, 0, 0, 0], [0, 0, 0, 0, 0]], dtype=uint8)
- cucim.skimage.morphology.medial_axis(image, mask=None, return_distance=False, *, random_state=None)#
Compute the medial axis transform of a binary image.
- Parameters
- imagebinary ndarray, shape (M, N)
The image of the shape to be skeletonized.
- maskbinary ndarray, shape (M, N), optional
If a mask is given, only those elements in image with a true value in mask are used for computing the medial axis.
- return_distancebool, optional
If true, the distance transform is returned as well as the skeleton.
- random_state{None, int, numpy.random.Generator}, optional
If random_state is None the numpy.random.Generator singleton is used. If random_state is an int, a new
Generatorinstance is used, seeded with random_state. If random_state is already aGeneratorinstance then that instance is used.New in version 0.19.
- Returns
- outndarray of bools
Medial axis transform of the image
- distndarray of ints, optional
Distance transform of the image (only returned if return_distance is True)
See also
skeletonize
Notes
This algorithm computes the medial axis transform of an image as the ridges of its distance transform.
- The different steps of the algorithm are as follows
A lookup table is used, that assigns 0 or 1 to each configuration of the 3x3 binary square, whether the central pixel should be removed or kept. We want a point to be removed if it has more than one neighbor and if removing it does not change the number of connected components.
The distance transform to the background is computed, as well as the cornerness of the pixel.
The foreground (value of 1) points are ordered by the distance transform, then the cornerness.
A cython function is called to reduce the image to its skeleton. It processes pixels in the order determined at the previous step, and removes or maintains a pixel according to the lookup table. Because of the ordering, it is possible to process all pixels in only one pass.
Examples
>>> square = np.zeros((7, 7), dtype=np.uint8) >>> square[1:-1, 2:-2] = 1 >>> square array([[0, 0, 0, 0, 0, 0, 0], [0, 0, 1, 1, 1, 0, 0], [0, 0, 1, 1, 1, 0, 0], [0, 0, 1, 1, 1, 0, 0], [0, 0, 1, 1, 1, 0, 0], [0, 0, 1, 1, 1, 0, 0], [0, 0, 0, 0, 0, 0, 0]], dtype=uint8) >>> medial_axis(square).astype(np.uint8) array([[0, 0, 0, 0, 0, 0, 0], [0, 0, 1, 0, 1, 0, 0], [0, 0, 0, 1, 0, 0, 0], [0, 0, 0, 1, 0, 0, 0], [0, 0, 0, 1, 0, 0, 0], [0, 0, 1, 0, 1, 0, 0], [0, 0, 0, 0, 0, 0, 0]], dtype=uint8)
- cucim.skimage.morphology.octagon(m, n, dtype=<class 'numpy.uint8'>, *, decomposition=None)#
Generates an octagon shaped footprint.
For a given size of (m) horizontal and vertical sides and a given (n) height or width of slanted sides octagon is generated. The slanted sides are 45 or 135 degrees to the horizontal axis and hence the widths and heights are equal. The overall size of the footprint along a single axis will be
m + 2 * n.- Parameters
- mint
The size of the horizontal and vertical sides.
- nint
The height or width of the slanted sides.
- Returns
- footprintcupy.ndarray
The footprint where elements of the neighborhood are 1 and 0 otherwise. When decomposition is None, this is just a numpy.ndarray. Otherwise, this will be a tuple whose length is equal to the number of unique structuring elements to apply (see Notes for more detail)
- Other Parameters
- dtypedata-type, optional
The data type of the footprint.
- decomposition{None, ‘sequence’}, optional
If None, a single array is returned. For ‘sequence’, a tuple of smaller footprints is returned. Applying this series of smaller footprints will given an identical result to a single, larger footprint, but with better computational performance. See Notes for more details.
Notes
When decomposition is not None, each element of the footprint tuple is a 2-tuple of the form
(ndarray, num_iter)that specifies a footprint array and the number of iterations it is to be applied.For either binary or grayscale morphology, using
decomposition='sequence'was observed to have a performance benefit, with the magnitude of the benefit increasing with increasing footprint size.
- cucim.skimage.morphology.octahedron(radius, dtype=<class 'numpy.uint8'>, *, decomposition=None)#
Generates a octahedron-shaped footprint.
This is the 3D equivalent of a diamond. A pixel is part of the neighborhood (i.e. labeled 1) if the city block/Manhattan distance between it and the center of the neighborhood is no greater than radius.
- Parameters
- radiusint
The radius of the octahedron-shaped footprint.
- Returns
- footprintcupy.ndarray
The footprint where elements of the neighborhood are 1 and 0 otherwise. When decomposition is None, this is just a numpy.ndarray. Otherwise, this will be a tuple whose length is equal to the number of unique structuring elements to apply (see Notes for more detail)
- Other Parameters
- dtypedata-type, optional
The data type of the footprint.
- decomposition{None, ‘sequence’}, optional
If None, a single array is returned. For ‘sequence’, a tuple of smaller footprints is returned. Applying this series of smaller footprints will given an identical result to a single, larger footprint, but with better computational performance. See Notes for more details.
Notes
When decomposition is not None, each element of the footprint tuple is a 2-tuple of the form
(ndarray, num_iter)that specifies a footprint array and the number of iterations it is to be applied.For either binary or grayscale morphology, using
decomposition='sequence'was observed to have a performance benefit, with the magnitude of the benefit increasing with increasing footprint size.
- cucim.skimage.morphology.opening(image, footprint=None, out=None)#
Return grayscale morphological opening of an image.
The morphological opening of an image is defined as an erosion followed by a dilation. Opening can remove small bright spots (i.e. “salt”) and connect small dark cracks. This tends to “open” up (dark) gaps between (bright) features.
- Parameters
- imagecupy.ndarray
Image array.
- footprintcupy.ndarray, optional
The neighborhood expressed as a 2-D array of 1’s and 0’s. If None, use a cross-shaped footprint (connectivity=1). The footprint can also be provided as a sequence of smaller footprints as described in the notes below.
- outcupy.ndarray, optional
The array to store the result of the morphology. If None is passed, a new array will be allocated.
- Returns
- openingcupy.ndarray, same shape and type as image
The result of the morphological opening.
- Other Parameters
- selemDEPRECATED
Deprecated in favor of footprint.
Deprecated since version 22.02.00.
Notes
The footprint can also be a provided as a sequence of 2-tuples where the first element of each 2-tuple is a footprint ndarray and the second element is an integer describing the number of times it should be iterated. For example
footprint=[(cp.ones((9, 1)), 1), (cp.ones((1, 9)), 1)]would apply a 9x1 footprint followed by a 1x9 footprint resulting in a net effect that is the same asfootprint=cp.ones((9, 9)), but with lower computational cost. Most of the builtin footprints such asskimage.morphology.diskprovide an option to automically generate a footprint sequence of this type.Examples
>>> # Open up gap between two bright regions (but also shrink regions) >>> import cupy as cp >>> from cucim.skimage.morphology import square >>> bad_connection = cp.asarray([[1, 0, 0, 0, 1], ... [1, 1, 0, 1, 1], ... [1, 1, 1, 1, 1], ... [1, 1, 0, 1, 1], ... [1, 0, 0, 0, 1]], dtype=cp.uint8) >>> opening(bad_connection, square(3)) array([[0, 0, 0, 0, 0], [1, 1, 0, 1, 1], [1, 1, 0, 1, 1], [1, 1, 0, 1, 1], [0, 0, 0, 0, 0]], dtype=uint8)
- cucim.skimage.morphology.reconstruction(seed, mask, method='dilation', footprint=None, offset=None)#
Perform a morphological reconstruction of an image.
Morphological reconstruction by dilation is similar to basic morphological dilation: high-intensity values will replace nearby low-intensity values. The basic dilation operator, however, uses a footprint to determine how far a value in the input image can spread. In contrast, reconstruction uses two images: a “seed” image, which specifies the values that spread, and a “mask” image, which gives the maximum allowed value at each pixel. The mask image, like the footprint, limits the spread of high-intensity values. Reconstruction by erosion is simply the inverse: low-intensity values spread from the seed image and are limited by the mask image, which represents the minimum allowed value.
Alternatively, you can think of reconstruction as a way to isolate the connected regions of an image. For dilation, reconstruction connects regions marked by local maxima in the seed image: neighboring pixels less-than-or-equal-to those seeds are connected to the seeded region. Local maxima with values larger than the seed image will get truncated to the seed value.
- Parameters
- seedndarray
The seed image (a.k.a. marker image), which specifies the values that are dilated or eroded.
- maskndarray
The maximum (dilation) / minimum (erosion) allowed value at each pixel.
- method{‘dilation’|’erosion’}, optional
Perform reconstruction by dilation or erosion. In dilation (or erosion), the seed image is dilated (or eroded) until limited by the mask image. For dilation, each seed value must be less than or equal to the corresponding mask value; for erosion, the reverse is true. Default is ‘dilation’.
- footprintndarray, optional
The neighborhood expressed as an n-D array of 1’s and 0’s. Default is the n-D square of radius equal to 1 (i.e. a 3x3 square for 2D images, a 3x3x3 cube for 3D images, etc.)
- offsetndarray, optional
The coordinates of the center of the footprint. Default is located on the geometrical center of the footprint, in that case footprint dimensions must be odd.
- Returns
- reconstructedndarray
The result of morphological reconstruction.
- Other Parameters
- selemDEPRECATED
Deprecated in favor of footprint.
Deprecated since version 22.02.00.
Notes
The algorithm is taken from [1]. Applications for grayscale reconstruction are discussed in [2] and [3].
References
- 1
Robinson, “Efficient morphological reconstruction: a downhill filter”, Pattern Recognition Letters 25 (2004) 1759-1767.
- 2
Vincent, L., “Morphological Grayscale Reconstruction in Image Analysis: Applications and Efficient Algorithms”, IEEE Transactions on Image Processing (1993)
- 3
Soille, P., “Morphological Image Analysis: Principles and Applications”, Chapter 6, 2nd edition (2003), ISBN 3540429883.
Examples
>>> import cupy as cp >>> from cucim.skimage.morphology import reconstruction
First, we create a sinusoidal mask image with peaks at middle and ends.
>>> x = cp.linspace(0, 4 * np.pi) >>> y_mask = cp.cos(x)
Then, we create a seed image initialized to the minimum mask value (for reconstruction by dilation, min-intensity values don’t spread) and add “seeds” to the left and right peak, but at a fraction of peak value (1).
>>> y_seed = y_mask.min() * cp.ones_like(x) >>> y_seed[0] = 0.5 >>> y_seed[-1] = 0 >>> y_rec = reconstruction(y_seed, y_mask)
The reconstructed image (or curve, in this case) is exactly the same as the mask image, except that the peaks are truncated to 0.5 and 0. The middle peak disappears completely: Since there were no seed values in this peak region, its reconstructed value is truncated to the surrounding value (-1).
As a more practical example, we try to extract the bright features of an image by subtracting a background image created by reconstruction.
>>> y, x = cp.mgrid[:20:0.5, :20:0.5] >>> bumps = cp.sin(x) + cp.sin(y)
To create the background image, set the mask image to the original image, and the seed image to the original image with an intensity offset, h.
>>> h = 0.3 >>> seed = bumps - h >>> background = reconstruction(seed, bumps)
The resulting reconstructed image looks exactly like the original image, but with the peaks of the bumps cut off. Subtracting this reconstructed image from the original image leaves just the peaks of the bumps
>>> hdome = bumps - background
This operation is known as the h-dome of the image and leaves features of height h in the subtracted image.
- cucim.skimage.morphology.rectangle(nrows, ncols, dtype=<class 'numpy.uint8'>, *, decomposition=None)#
Generates a flat, rectangular-shaped footprint.
Every pixel in the rectangle generated for a given width and given height belongs to the neighborhood.
- Parameters
- nrowsint
The number of rows of the rectangle.
- ncolsint
The number of columns of the rectangle.
- Returns
- footprintcupy.ndarray
A footprint consisting only of ones, i.e. every pixel belongs to the neighborhood. When decomposition is None, this is just a numpy.ndarray. Otherwise, this will be a tuple whose length is equal to the number of unique structuring elements to apply (see Notes for more detail)
- Other Parameters
- dtypedata-type, optional
The data type of the footprint.
- decomposition{None, ‘separable’, ‘sequence’}, optional
If None, a single array is returned. For ‘sequence’, a tuple of smaller footprints is returned. Applying this series of smaller footprints will given an identical result to a single, larger footprint, but often with better computational performance. See Notes for more details. With ‘separable’, this function uses separable 1D footprints for each axis. Whether ‘seqeunce’ or ‘separable’ is computationally faster may be architecture-dependent.
- heightDEPRECATED
Deprecated in favor of ncols.
Deprecated since version 21.06.00.
- widthDEPRECATED
Deprecated in favor of nrows.
Deprecated since version 21.06.00.
Notes
When decomposition is not None, each element of the footprint tuple is a 2-tuple of the form
(ndarray, num_iter)that specifies a footprint array and the number of iterations it is to be applied.For binary morphology, using
decomposition='sequence'was observed to give better performance, with the magnitude of the performance increase rapidly increasing with footprint size. For grayscale morphology with rectangular footprints, it is recommended to usedecomposition=Nonesince the internal SciPy functions that are called already have a fast implementation based on separable 1D sliding windows.The sequence decomposition mode only supports odd valued nrows and ncols. If either nrows or ncols is even, the sequence used will be identical to
decomposition='separable'.The use of
widthandheighthas been deprecated in version 0.18.0. Usenrowsandncolsinstead.
- cucim.skimage.morphology.remove_small_holes(ar, area_threshold=64, connectivity=1, in_place=False, *, out=None)#
Remove contiguous holes smaller than the specified size.
- Parameters
- arndarray (arbitrary shape, int or bool type)
The array containing the connected components of interest.
- area_thresholdint, optional (default: 64)
The maximum area, in pixels, of a contiguous hole that will be filled. Replaces min_size.
- connectivityint, {1, 2, …, ar.ndim}, optional (default: 1)
The connectivity defining the neighborhood of a pixel.
- in_placebool, optional (default: False)
If True, remove the connected components in the input array itself. Otherwise, make a copy. Deprecated since version 22.02.00. Please use out instead.
- outndarray
Array of the same shape as ar and bool dtype, into which the output is placed. By default, a new array is created.
- Returns
- outndarray, same shape and type as input ar
The input array with small holes within connected components removed.
- Raises
- TypeError
If the input array is of an invalid type, such as float or string.
- ValueError
If the input array contains negative values.
Notes
If the array type is int, it is assumed that it contains already-labeled objects. The labels are not kept in the output image (this function always outputs a bool image). It is suggested that labeling is completed after using this function.
Examples
>>> import cupy as cp >>> from cucim.skimage import morphology >>> a = cp.array([[1, 1, 1, 1, 1, 0], ... [1, 1, 1, 0, 1, 0], ... [1, 0, 0, 1, 1, 0], ... [1, 1, 1, 1, 1, 0]], bool) >>> b = morphology.remove_small_holes(a, 2) >>> b array([[ True, True, True, True, True, False], [ True, True, True, True, True, False], [ True, False, False, True, True, False], [ True, True, True, True, True, False]]) >>> c = morphology.remove_small_holes(a, 2, connectivity=2) >>> c array([[ True, True, True, True, True, False], [ True, True, True, False, True, False], [ True, False, False, True, True, False], [ True, True, True, True, True, False]]) >>> d = morphology.remove_small_holes(a, 2, out=a) >>> d is a True
- cucim.skimage.morphology.remove_small_objects(ar, min_size=64, connectivity=1, in_place=False, *, out=None)#
Remove objects smaller than the specified size.
Expects ar to be an array with labeled objects, and removes objects smaller than min_size. If ar is bool, the image is first labeled. This leads to potentially different behavior for bool and 0-and-1 arrays.
- Parameters
- arndarray (arbitrary shape, int or bool type)
The array containing the objects of interest. If the array type is int, the ints must be non-negative.
- min_sizeint, optional (default: 64)
The smallest allowable object size.
- connectivityint, {1, 2, …, ar.ndim}, optional (default: 1)
The connectivity defining the neighborhood of a pixel. Used during labelling if ar is bool.
- in_placebool, optional (default: False)
If
True, remove the objects in the input array itself. Otherwise, make a copy. Deprecated since version 22.02.00. Please use out instead.- outndarray
Array of the same shape as ar, into which the output is placed. By default, a new array is created.
- Returns
- outndarray, same shape and type as input ar
The input array with small connected components removed.
- Raises
- TypeError
If the input array is of an invalid type, such as float or string.
- ValueError
If the input array contains negative values.
Examples
>>> import cupy as cp >>> from cucim.skimage import morphology >>> a = cp.array([[0, 0, 0, 1, 0], ... [1, 1, 1, 0, 0], ... [1, 1, 1, 0, 1]], bool) >>> b = morphology.remove_small_objects(a, 6) >>> b array([[False, False, False, False, False], [ True, True, True, False, False], [ True, True, True, False, False]]) >>> c = morphology.remove_small_objects(a, 7, connectivity=2) >>> c array([[False, False, False, True, False], [ True, True, True, False, False], [ True, True, True, False, False]]) >>> d = morphology.remove_small_objects(a, 6, out=a) >>> d is a True
- cucim.skimage.morphology.square(width, dtype=<class 'numpy.uint8'>, *, decomposition=None)#
Generates a flat, square-shaped footprint.
Every pixel along the perimeter has a chessboard distance no greater than radius (radius=floor(width/2)) pixels.
- Parameters
- widthint
The width and height of the square.
- Returns
- footprintcupy.ndarray
The footprint where elements of the neighborhood are 1 and 0 otherwise. When decomposition is None, this is just a numpy.ndarray. Otherwise, this will be a tuple whose length is equal to the number of unique structuring elements to apply (see Notes for more detail)
- Other Parameters
- dtypedata-type, optional
The data type of the footprint.
- decomposition{None, ‘separable’, ‘sequence’}, optional
If None, a single array is returned. For ‘sequence’, a tuple of smaller footprints is returned. Applying this series of smaller footprints will given an identical result to a single, larger footprint, but often with better computational performance. See Notes for more details. With ‘separable’, this function uses separable 1D footprints for each axis. Whether ‘seqeunce’ or ‘separable’ is computationally faster may be architecture-dependent.
Notes
When decomposition is not None, each element of the footprint tuple is a 2-tuple of the form
(ndarray, num_iter)that specifies a footprint array and the number of iterations it is to be applied.For binary morphology, using
decomposition='sequence'ordecomposition='separable'were observed to give better performance thandecomposition=None, with the magnitude of the performance increase rapidly increasing with footprint size. For grayscale morphology with square footprints, it is recommended to usedecomposition=Nonesince the internal SciPy functions that are called already have a fast implementation based on separable 1D sliding windows.The ‘sequence’ decomposition mode only supports odd valued width. If width is even, the sequence used will be identical to the ‘separable’ mode.
- cucim.skimage.morphology.star(a, dtype=<class 'numpy.uint8'>)#
Generates a star shaped footprint.
Start has 8 vertices and is an overlap of square of size 2*a + 1 with its 45 degree rotated version. The slanted sides are 45 or 135 degrees to the horizontal axis.
- Parameters
- aint
Parameter deciding the size of the star structural element. The side of the square array returned is 2*a + 1 + 2*floor(a / 2).
- Returns
- footprintcupy.ndarray
The footprint where elements of the neighborhood are 1 and 0 otherwise.
- Other Parameters
- dtypedata-type, optional
The data type of the footprint.
- cucim.skimage.morphology.thin(image, max_num_iter=None)#
Perform morphological thinning of a binary image.
- Parameters
- imagebinary (M, N) ndarray
The image to be thinned.
- max_num_iterint, number of iterations, optional
Regardless of the value of this parameter, the thinned image is returned immediately if an iteration produces no change. If this parameter is specified it thus sets an upper bound on the number of iterations performed.
- Returns
- outndarray of bool
Thinned image.
- Other Parameters
- max_iterDEPRECATED
Deprecated in favor of max_num_iter.
Deprecated since version 22.02.00.
See also
Notes
This algorithm [1] works by making multiple passes over the image, removing pixels matching a set of criteria designed to thin connected regions while preserving eight-connected components and 2 x 2 squares [2]. In each of the two sub-iterations the algorithm correlates the intermediate skeleton image with a neighborhood mask, then looks up each neighborhood in a lookup table indicating whether the central pixel should be deleted in that sub-iteration.
References
- 1
Z. Guo and R. W. Hall, “Parallel thinning with two-subiteration algorithms,” Comm. ACM, vol. 32, no. 3, pp. 359-373, 1989. DOI:10.1145/62065.62074
- 2
Lam, L., Seong-Whan Lee, and Ching Y. Suen, “Thinning Methodologies-A Comprehensive Survey,” IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol 14, No. 9, p. 879, 1992. DOI:10.1109/34.161346
Examples
>>> square = np.zeros((7, 7), dtype=np.uint8) >>> square[1:-1, 2:-2] = 1 >>> square[0, 1] = 1 >>> square array([[0, 1, 0, 0, 0, 0, 0], [0, 0, 1, 1, 1, 0, 0], [0, 0, 1, 1, 1, 0, 0], [0, 0, 1, 1, 1, 0, 0], [0, 0, 1, 1, 1, 0, 0], [0, 0, 1, 1, 1, 0, 0], [0, 0, 0, 0, 0, 0, 0]], dtype=uint8) >>> skel = thin(square) >>> skel.astype(np.uint8) array([[0, 1, 0, 0, 0, 0, 0], [0, 0, 1, 0, 0, 0, 0], [0, 0, 0, 1, 0, 0, 0], [0, 0, 0, 1, 0, 0, 0], [0, 0, 0, 1, 0, 0, 0], [0, 0, 0, 0, 0, 0, 0], [0, 0, 0, 0, 0, 0, 0]], dtype=uint8)
- cucim.skimage.morphology.white_tophat(image, footprint=None, out=None)#
Return white top hat of an image.
The white top hat of an image is defined as the image minus its morphological opening. This operation returns the bright spots of the image that are smaller than the footprint.
- Parameters
- imagecupy.ndarray
Image array.
- footprintcupy.ndarray, optional
The neighborhood expressed as a 2-D array of 1’s and 0’s. If None, use a cross-shaped footprint (connectivity=1). The footprint can also be provided as a sequence of smaller footprints as described in the notes below.
- outcupy.ndarray, optional
The array to store the result of the morphology. If None is passed, a new array will be allocated.
- Returns
- outcupy.ndarray, same shape and type as image
The result of the morphological white top hat.
- Other Parameters
- selemDEPRECATED
Deprecated in favor of footprint.
Deprecated since version 22.02.00.
See also
Notes
The footprint can also be a provided as a sequence of 2-tuples where the first element of each 2-tuple is a footprint ndarray and the second element is an integer describing the number of times it should be iterated. For example
footprint=[(cp.ones((9, 1)), 1), (cp.ones((1, 9)), 1)]would apply a 9x1 footprint followed by a 1x9 footprint resulting in a net effect that is the same asfootprint=cp.ones((9, 9)), but with lower computational cost. Most of the builtin footprints such asskimage.morphology.diskprovide an option to automically generate a footprint sequence of this type.References
Examples
>>> # Subtract grey background from bright peak >>> import cupy as cp >>> from cucim.skimage.morphology import square >>> bright_on_grey = cp.asarray([[2, 3, 3, 3, 2], ... [3, 4, 5, 4, 3], ... [3, 5, 9, 5, 3], ... [3, 4, 5, 4, 3], ... [2, 3, 3, 3, 2]], dtype=cp.uint8) >>> white_tophat(bright_on_grey, square(3)) array([[0, 0, 0, 0, 0], [0, 0, 1, 0, 0], [0, 1, 5, 1, 0], [0, 0, 1, 0, 0], [0, 0, 0, 0, 0]], dtype=uint8)
registration#
- cucim.skimage.registration.optical_flow_ilk(reference_image, moving_image, *, radius=7, num_warp=10, gaussian=False, prefilter=False, dtype=<class 'numpy.float32'>)#
Coarse to fine optical flow estimator.
The iterative Lucas-Kanade (iLK) solver is applied at each level of the image pyramid. iLK [1] is a fast and robust alternative to TVL1 algorithm although less accurate for rendering flat surfaces and object boundaries (see [2]).
- Parameters
- reference_imagendarray, shape (M, N[, P[, …]])
The first gray scale image of the sequence.
- moving_imagendarray, shape (M, N[, P[, …]])
The second gray scale image of the sequence.
- radiusint, optional
Radius of the window considered around each pixel.
- num_warpint, optional
Number of times moving_image is warped.
- gaussianbool, optional
If True, a Gaussian kernel is used for the local integration. Otherwise, a uniform kernel is used.
- prefilterbool, optional
Whether to prefilter the estimated optical flow before each image warp. When True, a median filter with window size 3 along each axis is applied. This helps to remove potential outliers.
- dtypedtype, optional
Output data type: must be floating point. Single precision provides good results and saves memory usage and computation time compared to double precision.
- Returns
- flowndarray, shape ((reference_image.ndim, M, N[, P[, …]])
The estimated optical flow components for each axis.
Notes
The implemented algorithm is described in Table2 of [1].
Color images are not supported.
References
- 1(1,2)
Le Besnerais, G., & Champagnat, F. (2005, September). Dense optical flow by iterative local window registration. In IEEE International Conference on Image Processing 2005 (Vol. 1, pp. I-137). IEEE. DOI:10.1109/ICIP.2005.1529706
- 2
Plyer, A., Le Besnerais, G., & Champagnat, F. (2016). Massively parallel Lucas Kanade optical flow for real-time video processing applications. Journal of Real-Time Image Processing, 11(4), 713-730. DOI:10.1007/s11554-014-0423-0
Examples
>>> import cupy as cp >>> from skimage.data import stereo_motorcycle >>> from cucim.skimage.color import rgb2gray >>> from cucim.skimage.registration import optical_flow_ilk >>> reference_image, moving_image, disp = map(cp.array, stereo_motorcycle()) >>> # --- Convert the images to gray level: color is not supported. >>> reference_image = rgb2gray(reference_image) >>> moving_image = rgb2gray(moving_image) >>> flow = optical_flow_ilk(moving_image, reference_image)
- cucim.skimage.registration.optical_flow_tvl1(reference_image, moving_image, *, attachment=15, tightness=0.3, num_warp=5, num_iter=10, tol=0.0001, prefilter=False, dtype=<class 'numpy.float32'>)#
Coarse to fine optical flow estimator.
The TV-L1 solver is applied at each level of the image pyramid. TV-L1 is a popular algorithm for optical flow estimation introduced by Zack et al. [1], improved in [2] and detailed in [3].
- Parameters
- reference_imagendarray, shape (M, N[, P[, …]])
The first gray scale image of the sequence.
- moving_imagendarray, shape (M, N[, P[, …]])
The second gray scale image of the sequence.
- attachmentfloat, optional
Attachment parameter (\(\lambda\) in [1]). The smaller this parameter is, the smoother the returned result will be.
- tightnessfloat, optional
Tightness parameter (\(\tau\) in [1]). It should have a small value in order to maintain attachement and regularization parts in correspondence.
- num_warpint, optional
Number of times image1 is warped.
- num_iterint, optional
Number of fixed point iteration.
- tolfloat, optional
Tolerance used as stopping criterion based on the L² distance between two consecutive values of (u, v).
- prefilterbool, optional
Whether to prefilter the estimated optical flow before each image warp. When True, a median filter with window size 3 along each axis is applied. This helps to remove potential outliers.
- dtypedtype, optional
Output data type: must be floating point. Single precision provides good results and saves memory usage and computation time compared to double precision.
- Returns
- flowndarray, shape ((image0.ndim, M, N[, P[, …]])
The estimated optical flow components for each axis.
Notes
Color images are not supported.
References
- 1(1,2,3)
Zach, C., Pock, T., & Bischof, H. (2007, September). A duality based approach for realtime TV-L 1 optical flow. In Joint pattern recognition symposium (pp. 214-223). Springer, Berlin, Heidelberg. DOI:10.1007/978-3-540-74936-3_22
- 2
Wedel, A., Pock, T., Zach, C., Bischof, H., & Cremers, D. (2009). An improved algorithm for TV-L 1 optical flow. In Statistical and geometrical approaches to visual motion analysis (pp. 23-45). Springer, Berlin, Heidelberg. DOI:10.1007/978-3-642-03061-1_2
- 3
Pérez, J. S., Meinhardt-Llopis, E., & Facciolo, G. (2013). TV-L1 optical flow estimation. Image Processing On Line, 2013, 137-150. DOI:10.5201/ipol.2013.26
Examples
>>> import cupy as cp >>> from cucim.skimage.color import rgb2gray >>> from skimage.data import stereo_motorcycle >>> from cucim.skimage.registration import optical_flow_tvl1 >>> image0, image1, disp = [cp.array(a) for a in stereo_motorcycle()] >>> # --- Convert the images to gray level: color is not supported. >>> image0 = rgb2gray(image0) >>> image1 = rgb2gray(image1) >>> flow = optical_flow_tvl1(image1, image0)
- cucim.skimage.registration.phase_cross_correlation(reference_image, moving_image, *, upsample_factor=1, space='real', return_error=True, reference_mask=None, moving_mask=None, overlap_ratio=0.3, normalization='phase')#
Efficient subpixel image translation registration by cross-correlation.
This code gives the same precision as the FFT upsampled cross-correlation in a fraction of the computation time and with reduced memory requirements. It obtains an initial estimate of the cross-correlation peak by an FFT and then refines the shift estimation by upsampling the DFT only in a small neighborhood of that estimate by means of a matrix-multiply DFT [1].
- Parameters
- reference_imagearray
Reference image.
- moving_imagearray
Image to register. Must be same dimensionality as
reference_image.- upsample_factorint, optional
Upsampling factor. Images will be registered to within
1 / upsample_factorof a pixel. For exampleupsample_factor == 20means the images will be registered within 1/20th of a pixel. Default is 1 (no upsampling). Not used if any ofreference_maskormoving_maskis not None.- spacestring, one of “real” or “fourier”, optional
Defines how the algorithm interprets input data. “real” means data will be FFT’d to compute the correlation, while “fourier” data will bypass FFT of input data. Case insensitive. Not used if any of
reference_maskormoving_maskis not None.- return_errorbool, optional
Returns error and phase difference if on, otherwise only shifts are returned. Has noeffect if any of
reference_maskormoving_maskis not None. In this case only shifts is returned.- reference_maskndarray
Boolean mask for
reference_image. The mask should evaluate toTrue(or 1) on valid pixels.reference_maskshould have the same shape asreference_image.- moving_maskndarray or None, optional
Boolean mask for
moving_image. The mask should evaluate toTrue(or 1) on valid pixels.moving_maskshould have the same shape asmoving_image. IfNone,reference_maskwill be used.- overlap_ratiofloat, optional
Minimum allowed overlap ratio between images. The correlation for translations corresponding with an overlap ratio lower than this threshold will be ignored. A lower overlap_ratio leads to smaller maximum translation, while a higher overlap_ratio leads to greater robustness against spurious matches due to small overlap between masked images. Used only if one of
reference_maskormoving_maskis None.- normalization{“phase”, None}
The type of normalization to apply to the cross-correlation. This parameter is unused when masks (reference_mask and moving_mask) are supplied.
- Returns
- shiftsndarray
Shift vector (in pixels) required to register
moving_imagewithreference_image. Axis ordering is consistent with numpy (e.g. Z, Y, X)- errorfloat
Translation invariant normalized RMS error between
reference_imageandmoving_image.- phasedifffloat
Global phase difference between the two images (should be zero if images are non-negative).
Notes
The use of cross-correlation to estimate image translation has a long history dating back to at least [2]. The “phase correlation” method (selected by
normalization="phase") was first proposed in [3]. Publications [1] and [2] use an unnormalized cross-correlation (normalization=None). Which form of normalization is better is application-dependent. For example, the phase correlation method works well in registering images under different illumination, but is not very robust to noise. In a high noise scenario, the unnormalized method may be preferable.When masks are provided, a masked normalized cross-correlation algorithm is used [5], [6].
References
- 1(1,2)
Manuel Guizar-Sicairos, Samuel T. Thurman, and James R. Fienup, “Efficient subpixel image registration algorithms,” Optics Letters 33, 156-158 (2008). DOI:10.1364/OL.33.000156
- 2(1,2)
P. Anuta, Spatial registration of multispectral and multitemporal digital imagery using fast Fourier transform techniques, IEEE Trans. Geosci. Electron., vol. 8, no. 4, pp. 353–368, Oct. 1970. DOI:10.1109/TGE.1970.271435.
- 3
C. D. Kuglin D. C. Hines. The phase correlation image alignment method, Proceeding of IEEE International Conference on Cybernetics and Society, pp. 163-165, New York, NY, USA, 1975, pp. 163–165.
- 4
James R. Fienup, “Invariant error metrics for image reconstruction” Optics Letters 36, 8352-8357 (1997). DOI:10.1364/AO.36.008352
- 5
Dirk Padfield. Masked Object Registration in the Fourier Domain. IEEE Transactions on Image Processing, vol. 21(5), pp. 2706-2718 (2012). DOI:10.1109/TIP.2011.2181402
- 6
D. Padfield. “Masked FFT registration”. In Proc. Computer Vision and Pattern Recognition, pp. 2918-2925 (2010). DOI:10.1109/CVPR.2010.5540032
restoration#
- cucim.skimage.restoration.calibrate_denoiser(image, denoise_function, denoise_parameters, *, stride=4, approximate_loss=True, extra_output=False)#
Calibrate a denoising function and return optimal J-invariant version.
The returned function is partially evaluated with optimal parameter values set for denoising the input image.
- Parameters
- imagendarray
Input data to be denoised (converted using img_as_float).
- denoise_functionfunction
Denoising function to be calibrated.
- denoise_parametersdict of list
Ranges of parameters for denoise_function to be calibrated over.
- strideint, optional
Stride used in masking procedure that converts denoise_function to J-invariance.
- approximate_lossbool, optional
Whether to approximate the self-supervised loss used to evaluate the denoiser by only computing it on one masked version of the image. If False, the runtime will be a factor of stride**image.ndim longer.
- extra_outputbool, optional
If True, return parameters and losses in addition to the calibrated denoising function
- Returns
- best_denoise_functionfunction
The optimal J-invariant version of denoise_function.
- If extra_output is True, the following tuple is also returned:
- (parameters_tested, losses)tuple (list of dict, list of int)
List of parameters tested for denoise_function, as a dictionary of kwargs Self-supervised loss for each set of parameters in parameters_tested.
Notes
The calibration procedure uses a self-supervised mean-square-error loss to evaluate the performance of J-invariant versions of denoise_function. The minimizer of the self-supervised loss is also the minimizer of the ground-truth loss (i.e., the true MSE error) [1]. The returned function can be used on the original noisy image, or other images with similar characteristics.
- Increasing the stride increases the performance of best_denoise_function
at the expense of increasing its runtime. It has no effect on the runtime of the calibration.
References
- 1
J. Batson & L. Royer. Noise2Self: Blind Denoising by Self-Supervision, International Conference on Machine Learning, p. 524-533 (2019).
Examples
>>> import cupy as cp >>> from cucim.skimage import color >>> from skimage import data >>> from cucim.skimage.restoration import (denoise_tv_chambolle, ... calibrate_denoiser) >>> img = color.rgb2gray(cp.array(data.astronaut()[:50, :50])) >>> noisy = img + 0.5 * img.std() * cp.random.randn(*img.shape) >>> parameters = {'weight': cp.arange(0.01, 0.5, 0.05)} >>> denoising_function = calibrate_denoiser(noisy, denoise_tv_chambolle, ... denoise_parameters=parameters) >>> denoised_img = denoising_function(img)
- cucim.skimage.restoration.denoise_tv_chambolle(image, weight=0.1, eps=0.0002, max_num_iter=200, multichannel=False, *, channel_axis=None)#
Perform total-variation denoising on n-dimensional images.
- Parameters
- imagendarray of ints, uints or floats
Input data to be denoised. image can be of any numeric type, but it is cast into an ndarray of floats for the computation of the denoised image.
- weightfloat, optional
Denoising weight. The greater weight, the more denoising (at the expense of fidelity to input).
- epsfloat, optional
Relative difference of the value of the cost function that determines the stop criterion. The algorithm stops when:
(E_(n-1) - E_n) < eps * E_0
- max_num_iterint, optional
Maximal number of iterations used for the optimization.
- multichannelbool, optional
Apply total-variation denoising separately for each channel. This option should be true for color images, otherwise the denoising is also applied in the channels dimension. This argument is deprecated: specify channel_axis instead.
- channel_axisint or None, optional
If None, the image is assumed to be a grayscale (single channel) image. Otherwise, this parameter indicates which axis of the array corresponds to channels.
- Returns
- outndarray
Denoised image.
- Other Parameters
- multichannelDEPRECATED
Deprecated in favor of channel_axis.
Deprecated since version 22.02.00.
- n_iter_maxDEPRECATED
Deprecated in favor of max_num_iter.
Deprecated since version 22.06.00.
Notes
Make sure to set the multichannel parameter appropriately for color images.
The principle of total variation denoising is explained in https://en.wikipedia.org/wiki/Total_variation_denoising
The principle of total variation denoising is to minimize the total variation of the image, which can be roughly described as the integral of the norm of the image gradient. Total variation denoising tends to produce “cartoon-like” images, that is, piecewise-constant images.
This code is an implementation of the algorithm of Rudin, Fatemi and Osher that was proposed by Chambolle in [1].
References
- 1
A. Chambolle, An algorithm for total variation minimization and applications, Journal of Mathematical Imaging and Vision, Springer, 2004, 20, 89-97.
Examples
2D example on astronaut image:
>>> import cupy as cp >>> from cucim.skimage import color >>> from skimage import data >>> img = color.rgb2gray(cp.array(data.astronaut()[:50, :50])) >>> img += 0.5 * img.std() * cp.random.randn(*img.shape) >>> denoised_img = denoise_tv_chambolle(img, weight=60)
3D example on synthetic data:
>>> x, y, z = cp.ogrid[0:20, 0:20, 0:20] >>> mask = (x - 22)**2 + (y - 20)**2 + (z - 17)**2 < 8**2 >>> mask = mask.astype(float) >>> mask += 0.2*cp.random.randn(*mask.shape) >>> res = denoise_tv_chambolle(mask, weight=100)
- cucim.skimage.restoration.richardson_lucy(image, psf, num_iter=50, clip=True, filter_epsilon=None)#
Richardson-Lucy deconvolution.
- Parameters
- imagendarray
Input degraded image (can be N dimensional).
- psfndarray
The point spread function.
- num_iterint, optional
Number of iterations. This parameter plays the role of regularisation.
- clipboolean, optional
True by default. If true, pixel value of the result above 1 or under -1 are thresholded for skimage pipeline compatibility.
- filter_epsilon: float, optional
Value below which intermediate results become 0 to avoid division by small numbers.
- Returns
- im_deconvndarray
The deconvolved image.
- Other Parameters
- iterationsDEPRECATED
Deprecated in favor of num_iter.
Deprecated since version 22.02.00.
References
Examples
>>> import cupy as cp >>> from cucim.skimage import img_as_float, restoration >>> from skimage import data >>> camera = cp.asarray(img_as_float(cp.array(data.camera()))) >>> from cupyx.scipy.signal import convolve2d >>> psf = cp.ones((5, 5)) / 25 >>> camera = convolve2d(camera, psf, 'same') >>> camera += 0.1 * camera.std() * cp.random.standard_normal(camera.shape) >>> deconvolved = restoration.richardson_lucy(camera, psf, 5)
- cucim.skimage.restoration.unsupervised_wiener(image, psf, reg=None, user_params=None, is_real=True, clip=True, *, random_state=None)#
Unsupervised Wiener-Hunt deconvolution.
Return the deconvolution with a Wiener-Hunt approach, where the hyperparameters are automatically estimated. The algorithm is a stochastic iterative process (Gibbs sampler) described in the reference below. See also
wienerfunction.- Parameters
- image(M, N) ndarray
The input degraded image.
- psfndarray
The impulse response (input image’s space) or the transfer function (Fourier space). Both are accepted. The transfer function is automatically recognized as being complex (
cupy.iscomplexobj(psf)).- regndarray, optional
The regularisation operator. The Laplacian by default. It can be an impulse response or a transfer function, as for the psf.
- user_paramsdict, optional
Dictionary of parameters for the Gibbs sampler. See below.
- clipboolean, optional
True by default. If true, pixel values of the result above 1 or under -1 are thresholded for skimage pipeline compatibility.
- random_state{None, int, cupy.random.Generator}, optional
If random_state is None the cupy.random.Generator singleton is used. If random_state is an int, a new
Generatorinstance is used, seeded with random_state. If random_state is already aGeneratorinstance then that instance is used.
- Returns
- x_postmean(M, N) ndarray
The deconvolved image (the posterior mean).
- chainsdict
The keys
noiseandpriorcontain the chain list of noise and prior precision respectively.
- Other Parameters
- The keys of ``user_params`` are:
- thresholdfloat
The stopping criterion: the norm of the difference between to successive approximated solution (empirical mean of object samples, see Notes section). 1e-4 by default.
- burninint
The number of sample to ignore to start computation of the mean. 15 by default.
- min_num_iterint
The minimum number of iterations. 30 by default.
- max_num_iterint
The maximum number of iterations if
thresholdis not satisfied. 200 by default.- callbackcallable (None by default)
A user provided callable to which is passed, if the function exists, the current image sample for whatever purpose. The user can store the sample, or compute other moments than the mean. It has no influence on the algorithm execution and is only for inspection.
Notes
The estimated image is design as the posterior mean of a probability law (from a Bayesian analysis). The mean is defined as a sum over all the possible images weighted by their respective probability. Given the size of the problem, the exact sum is not tractable. This algorithm use of MCMC to draw image under the posterior law. The practical idea is to only draw highly probable images since they have the biggest contribution to the mean. At the opposite, the less probable images are drawn less often since their contribution is low. Finally the empirical mean of these samples give us an estimation of the mean, and an exact computation with an infinite sample set.
References
- 1
François Orieux, Jean-François Giovannelli, and Thomas Rodet, “Bayesian estimation of regularization and point spread function parameters for Wiener-Hunt deconvolution”, J. Opt. Soc. Am. A 27, 1593-1607 (2010)
https://www.osapublishing.org/josaa/abstract.cfm?URI=josaa-27-7-1593
Examples
>>> import cupy as cp >>> import cupyx.scipy.ndimage as ndi >>> from cucim.skimage import color, restoration >>> from skimage import data >>> img = color.rgb2gray(cp.array(data.astronaut())) >>> psf = cp.ones((5, 5)) / 25 >>> img = ndi.uniform_filter(img, size=psf.shape) >>> img += 0.1 * img.std() * cp.random.standard_normal(img.shape) >>> deconvolved_img = restoration.unsupervised_wiener(img, psf)
- cucim.skimage.restoration.wiener(image, psf, balance, reg=None, is_real=True, clip=True)#
Wiener-Hunt deconvolution
Return the deconvolution with a Wiener-Hunt approach (i.e. with Fourier diagonalisation).
- Parameters
- image(M, N) ndarray
Input degraded image
- psfndarray
Point Spread Function. This is assumed to be the impulse response (input image space) if the data-type is real, or the transfer function (Fourier space) if the data-type is complex. There is no constraints on the shape of the impulse response. The transfer function must be of shape (M, N) if is_real is True, (M, N // 2 + 1) otherwise (see cupy.fft.rfftn).
- balancefloat
The regularisation parameter value that tunes the balance between the data adequacy that improve frequency restoration and the prior adequacy that reduce frequency restoration (to avoid noise artifacts).
- regndarray, optional
The regularisation operator. The Laplacian by default. It can be an impulse response or a transfer function, as for the psf. Shape constraint is the same as for the psf parameter.
- is_realboolean, optional
True by default. Specify if
psfandregare provided with hermitian hypothesis, that is only half of the frequency plane is provided (due to the redundancy of Fourier transform of real signal). It’s apply only ifpsfand/orregare provided as transfer function. For the hermitian property seeuftmodule orcupy.fft.rfftn.- clipboolean, optional
True by default. If True, pixel values of the result above 1 or under -1 are thresholded for skimage pipeline compatibility.
- Returns
- im_deconv(M, N) ndarray
The deconvolved image.
Notes
This function applies the Wiener filter to a noisy and degraded image by an impulse response (or PSF). If the data model is
\[y = Hx + n\]where \(n\) is noise, \(H\) the PSF and \(x\) the unknown original image, the Wiener filter is
\[\hat x = F^\dagger (|\Lambda_H|^2 + \lambda |\Lambda_D|^2) \Lambda_H^\dagger F y\]where \(F\) and \(F^\dagger\) are the Fourier and inverse Fourier transforms respectively, \(\Lambda_H\) the transfer function (or the Fourier transform of the PSF, see [Hunt] below) and \(\Lambda_D\) the filter to penalize the restored image frequencies (Laplacian by default, that is penalization of high frequency). The parameter \(\lambda\) tunes the balance between the data (that tends to increase high frequency, even those coming from noise), and the regularization.
These methods are then specific to a prior model. Consequently, the application or the true image nature must corresponds to the prior model. By default, the prior model (Laplacian) introduce image smoothness or pixel correlation. It can also be interpreted as high-frequency penalization to compensate the instability of the solution with respect to the data (sometimes called noise amplification or “explosive” solution).
Finally, the use of Fourier space implies a circulant property of \(H\), see [Hunt].
References
- 1
François Orieux, Jean-François Giovannelli, and Thomas Rodet, “Bayesian estimation of regularization and point spread function parameters for Wiener-Hunt deconvolution”, J. Opt. Soc. Am. A 27, 1593-1607 (2010)
https://www.osapublishing.org/josaa/abstract.cfm?URI=josaa-27-7-1593
- 2
B. R. Hunt “A matrix theory proof of the discrete convolution theorem”, IEEE Trans. on Audio and Electroacoustics, vol. au-19, no. 4, pp. 285-288, dec. 1971
Examples
>>> import cupy as cp >>> import cupyx.scipy.ndimage as ndi >>> from cucim.skimage import color, restoration >>> from skimage import data >>> img = color.rgb2gray(cp.array(data.astronaut())) >>> psf = cp.ones((5, 5)) / 25 >>> img = ndi.uniform_filter(img, size=psf.shape) >>> img += 0.1 * img.std() * cp.random.standard_normal(