Surprisingly large variation in objects

We receive and test a variety of samples, and there is quite a large variation in the quality of the products.

For example, 0.1% distortion is 1 pixel distortion in 1000 pixels. 1 million pixel camera is 1000 x 1000 pixels, so 0.1% distortion will affect 1 pixel. in trying to find 1~2 pixel defects, it is a level that can never be ignored.

This is also why simple comparative inspection (pattern matching, image differencing, etc.) does not work as well as it should. The main point is how to consider the variation of objects.