Finally a smart exact match image finder with accurate results. Current image match operators use a lazy approach to find similar images. They scan the images and look for exact replica patterns in other images, regardless of content. The result is that they sometimes , or very often, get fooled by exact patterns that have nothing to do with the original image . Sometimes we are left to wonder.
This project from Carnegie Mellon University  uses a more human approach. It first compares the image with a pool of images that have nothing to do with the original. The reason ? To find out what is different thus notable. It then takes that information and proceeds with the match search. So, not only it searches for exact patterns, it also keeps on track by making sure that the notable element(s) of the image are included, discarding others.
The result is a much more efficient result, matching more precisely what a human would be looking for.
See video here:

The issue with image search on the internet is, however, not solved. With millions of images being uploaded everyday, with billions of websites changing their content daily, it is currently impossible to index and search all images. Sites like Tineye, although claiming 2 billion images indexed is still only indexing a nugget of the image universe. Not even  mighty Google can keep up.
Thus, this technology is certainly a step in the right direction and could work wonders in a closed image database but it still will not find all the images out there.