The scikit-image implementation of GLCM computes the grey level co-occurrence matrix for entire images or image patches. For some image labeling problems it is preferred to calculate GLCM features for each individual pixel using a sliding window approach as described in http://www.fp.ucalgary.ca/mhallbey/texture_calculations.htm. This approach can be implemented in scikit-image using a combination of view_as_windows, greycomatrix and greycoprops although I found it to be too slow for practical use. 

To overcome this I extended scikit-image with a function that performs the sliding window as fast cython code. It takes an image, distance, angle and window size and generates a texture image with the GLCM features ['mean','variance','contrast','dissimilarity','homogeneity','ASM','entropy','correlation'] calculated at every pixel. 

Would this be something people are interested in including as part of scikit-image?