Depends on how varied your images are.
- If it's any possible kind of building (wood, glass, steel, masonry, etc.) and any possible kind of masonry (brick, blocks, marble, limestone, stucco, cement, tile, different colors, etc.) then I think it could be difficult. You'd probably have to use SegNet to train a deep learning model with thousands of images.
- If you have a more limited set of images, like houses made of only red brick, you could possibly get by with traditional methods like a texture filter like stdfilt() or entropyfilt().
