Multi-dimensional image processing

CuPy provides multi-dimensional image processing functions. It supports a subset of scipy.ndimage interface.

Filters

cupyx.scipy.ndimage.convolve	Multi-dimensional convolution.
cupyx.scipy.ndimage.convolve1d	One-dimensional convolution.
cupyx.scipy.ndimage.correlate	Multi-dimensional correlate.
cupyx.scipy.ndimage.correlate1d	One-dimensional correlate.
cupyx.scipy.ndimage.gaussian_filter	Multi-dimensional Gaussian filter.
cupyx.scipy.ndimage.gaussian_filter1d	One-dimensional Gaussian filter along the given axis.
cupyx.scipy.ndimage.gaussian_gradient_magnitude	Multi-dimensional gradient magnitude using Gaussian derivatives.
cupyx.scipy.ndimage.gaussian_laplace	Multi-dimensional Laplace filter using Gaussian second derivatives.
cupyx.scipy.ndimage.generic_filter	Compute a multi-dimensional filter using the provided raw kernel or reduction kernel.
cupyx.scipy.ndimage.generic_filter1d	Compute a 1D filter along the given axis using the provided raw kernel.
cupyx.scipy.ndimage.generic_gradient_magnitude	Multi-dimensional gradient magnitude filter using a provided derivative function.
cupyx.scipy.ndimage.generic_laplace	Multi-dimensional Laplace filter using a provided second derivative function.
cupyx.scipy.ndimage.laplace	Multi-dimensional Laplace filter based on approximate second derivatives.
cupyx.scipy.ndimage.maximum_filter	Multi-dimensional maximum filter.
cupyx.scipy.ndimage.maximum_filter1d	Compute the maximum filter along a single axis.
cupyx.scipy.ndimage.median_filter	Multi-dimensional median filter.
cupyx.scipy.ndimage.minimum_filter	Multi-dimensional minimum filter.
cupyx.scipy.ndimage.minimum_filter1d	Compute the minimum filter along a single axis.
cupyx.scipy.ndimage.percentile_filter	Multi-dimensional percentile filter.
cupyx.scipy.ndimage.prewitt	Compute a Prewitt filter along the given axis.
cupyx.scipy.ndimage.rank_filter	Multi-dimensional rank filter.
cupyx.scipy.ndimage.sobel	Compute a Sobel filter along the given axis.
cupyx.scipy.ndimage.uniform_filter	Multi-dimensional uniform filter.
cupyx.scipy.ndimage.uniform_filter1d	One-dimensional uniform filter along the given axis.

Fourier Filters

cupyx.scipy.ndimage.fourier_gaussian	Multidimensional Gaussian shift filter.
cupyx.scipy.ndimage.fourier_shift	Multidimensional Fourier shift filter.
cupyx.scipy.ndimage.fourier_uniform	Multidimensional uniform shift filter.

Interpolation

cupyx.scipy.ndimage.affine_transform	Apply an affine transformation.
cupyx.scipy.ndimage.map_coordinates	Map the input array to new coordinates by interpolation.
cupyx.scipy.ndimage.rotate	Rotate an array.
cupyx.scipy.ndimage.shift	Shift an array.
cupyx.scipy.ndimage.zoom	Zoom an array.

Measurements

cupyx.scipy.ndimage.label	Labels features in an array.
cupyx.scipy.ndimage.mean	Calculates the mean of the values of an n-D image array, optionally
cupyx.scipy.ndimage.standard_deviation	Calculates the standard deviation of the values of an n-D image array, optionally at specified sub-regions.
cupyx.scipy.ndimage.sum	Calculates the sum of the values of an n-D image array, optionally
cupyx.scipy.ndimage.variance	Calculates the variance of the values of an n-D image array, optionally at specified sub-regions.

Morphology

cupyx.scipy.ndimage.grey_closing	Calculates a multi-dimensional greyscale closing.
cupyx.scipy.ndimage.grey_dilation	Calculates a greyscale dilation.
cupyx.scipy.ndimage.grey_erosion	Calculates a greyscale erosion.
cupyx.scipy.ndimage.grey_opening	Calculates a multi-dimensional greyscale opening.

OpenCV mode

cupyx.scipy.ndimage supports additional mode, opencv. If it is given, the function performs like cv2.warpAffine or cv2.resize. Example:

import cupyx.scipy.ndimage
import cupy as cp
import cv2

im = cv2.imread('TODO') # pls fill in your image path

trans_mat = cp.eye(4)
trans_mat[0][0] = trans_mat[1][1] = 0.5

smaller_shape = (im.shape[0] // 2, im.shape[1] // 2, 3)
smaller = cp.zeros(smaller_shape) # preallocate memory for resized image

cupyx.scipy.ndimage.affine_transform(im, trans_mat, output_shape=smaller_shape,
                                     output=smaller, mode='opencv')

cv2.imwrite('smaller.jpg', cp.asnumpy(smaller)) # smaller image saved locally