There is a nice online demo of a "Visual Google" here:

http://www.robots.ox.ac.uk/~vgg/research/vgoogle/

You give it a picture, and it returns all the frames in a movie that contain the picture. It works very well. The clever bit is that it uses a kind of text-retrieval inspired method to do it, so the processing time is essentially zero (just an inverted index lookup, with "visual words")