detecting near-duplicate images

edit

Does OpenCV have algorithms for duplicate image detection? By that, I mean finding in a collection of images sets of images that differ from some original by simple transformations such as compression (eg: to smaller pixel dimensions), minor cropping, changing to greyscale, minor colour balance changes etc..

Typical steps might involve:

• making feature representations of images (eg: SURF, SIFT etc..)
• matching features between pairs of images
• identifying sets of matching features as indicating a duplicate

For the first two, there are feature extractors and matchers. I am less sure about how to achieve the 3rd. Perhaps finding an affine transformation with a confidence level?