[FEA] add graycomatrix and graycoprops in skimage.feature

[maxbut38]

New generation of artificial intelligence (CNN or Classification) for texture analysis use the grey level co-occurence matrix (GLCM) and its features (contrast, dissimilarity, homogeneity, energy,correlation, ASM) for better classification.

Those methods are alreday in skimage.feature. Thus, i would know if it is possible to add to cucim.skimage.feature to make calculations faster.

Hereafter, the code to transform.

glcm = graycomatrix(image, distances=distances, 
			angles=angles, levels=levels,symmetric=False,normed=True)
contrast = graycoprops(glcm, 'contrast')
dissimilarity = graycoprops(glcm, 'dissimilarity')
homogeneity = graycoprops(glcm, 'homogeneity')
energy = graycoprops(glcm, 'energy')
correlation = graycoprops(glcm, 'correlation')
asm = graycoprops(glcm, 'ASM')
features = np.concatenate([contrast, dissimilarity, homogeneity, energy, correlation, asm])

[talakhlaif]

I would like to work on this issue and implement the graycomatrix

[monzelr]

I will help @talakhlaif with the implementation

[grlee77]

Thank you for volunteering to work on the feature.

I had briefly looked for existing implementations and found these two that are MIT-licensed:
https://github.com/Eve-ning/glcm-cupy
https://github.com/teamsar/glcmwithcuda

I did not try either one to see what the relative performance was or how close they are to the scikit-image API. If you have experience with either I would be glad to hear about it. The first one is already CuPy based and actually seems to optionally use cuCIM here:
https://github.com/Eve-ning/glcm-cupy/blob/master/glcm_cupy/cross/glcm_cross.py#L12