See attached pictures. They are all similar but clearly there are two sets, each is derived from simply impose a logo on top of its root image.
Your job is to find them out.
What you need to do are:
1. do an "exact" (to be further clarified below) pairwise match of two pictures and identify area of exact match and no match. For the example pictures, for group of pictures from the same root picture, this shall highlight CONCENTRATED areas of added logos (as non-match). For pictures from different root images but the same family, this shall also highlight CONCENTRATED areas of logos (as match).
2. apply the above algorithm to a given group of images from the same vendor (example see https://item.jd.com/12801479644.html and https://detail.tmall.com/item.htm?spm=a220m.1000858.1000725.1.d8117971DaUyl&id=527553220074&skuId=3565498258051&areaId=110100&user_id=1836892878&cat_id=2&is_b=1&rn=13840991599672ba37da35eb9c401f61#). We are expected to extract logo images precisely.
3. apply the same algorithm to a see of images, we shall be able to find all images derived from the same root image, together with identification of the root image, through pairwise comparisons. Example pictures can be found here:
4. "exact" here is defined as invariant with scaling, cropping, and color transformation, together with distortion due to image format conversion (loss due to png to jpg conversion, for example).