I tried PHOT (suggested by @applesoup) using the unofficial implementation given under https://github.com/thinkng but it did not work for these image sets. Maybe one needs to further investigate or tweak the algorithm a bit.
First, if you have sufficient amount of data, I do not believe that one could easily outperform a good deep architecture in the task ...
If I understand correctly, the question is, given many images which are result of different Edge Filter applied on the same image, how to actually mark edges.
Well, you basically created 25 tests for each pixel to decide whether or not it is an edge.
You could apply many approaches to decide:
Majority Votes - If more than half of the voters decided it ...
A periodic geometrical texture may appear as a localized wiggling pattern, with dominant orientations and 2D frequencies. Its correlation with Gabor atoms is bigger when the atom has orientation and frequencies close to the texture's, thus yielding feature vectors allowing texture detection or classification.
Looking at these references, this is what I would understand:
The Haralick descriptors only create a single value for each GLCM (as you also thought).
The Rampun paper creates 32 images, describing 32 different features for each input image.
The crucial point is this sentence: On the other hand, we used
a small window size of 5 x 5 throughout the process