Irregular pattern detection

Hi, I am still working with my project of detecting scratches and unpolished areas of metallic objects.
I’ve found some interesting solutions and get new ideas, but there is one problem.
Is this possible to found, detect or higlight the area that is clearly different from rest of the object? I mean with some operations on the image I can get the structure of material. The polished metal have the vertical lines that are identical to the movement of the robot. So the unpolished area or scratches are just simply overlapping this nice pattern.
Have you got any idea if it’s possible by using classic machine vision to classify these sections or it’s more complex? I’ve seen that OpenCv have Local Binary Pattern function, but I don’t know if it can handle non regular patterns.
The neural networks that I’m looking at can be tricky with this, because the surface is’nt flat and the elements can have multiple sizes and shapes.

As you can see on the image above, every defects are visible as the groups of white points.