Perceptual hashing could have been the technique used. For example, see: https://github.com/JohannesBuchner/imagehash

I use this for deduplicating my, er, 'home movies' collection.

EDIT: Here's a good explanation of one specific technique that should give you the general idea: http://www.hackerfactor.com/blog/?/archives/432-Looks-Like-I...