detection of paint defect in car door handles

edit

can you add an image without the red circle?

There is no out-of-the-box solution for this, but google scholar already provides tons of possible solutions!

Well, I think that if someone posts a question on a technical forum as it is, he deserves something different than 'try googling'. That said, my two cents:

• Regarding lighing you can try to move to a different colospace. For example you can try to normalize luminance of your image converting into a YUV colorspace, then making a histogram equalization on the Y channel, then converting it back to RGB;
• Regarding defect detection, I would try conversion to grayscale, contrast stretching (histogram), binary threshold and find contours with some filtering of results based on dimensions of your defect.

Euhm wait @procton, that is certainly true, and what I said is not, go search on google, but just a quick link to several scientific publications that tackle this problem and that can provide a solution?

I read most of the academic articles. But the articles are working on sections where the error is(They do not work on the whole thing). These photos were ideal. In fact, there will be other items behind the door arm.