Image search has always fascinated me, from large database pattern matching, (You draw a doodle of the image you are searching for with your mouse on a slate) to face recognizing for security purposes. Both the subjects are bordering each other as well as my latest fascination with medical imaging with pattern matching. But the Google researchers may look at it from another angle.
So how does this work?
The algorithm, (which is also presented in very technical and theoretical paper, which is very interesting if your field is image processing.) is to collect a bunch of images based on the search and then analyze the results to select best possible image among them.
I will be reading both the papers that were presented by Official Google Blog article, Visualrank,
Pagerank for product image search and YouTube datamining (which I have read before).
The see the Google blog, click on the link below.
Official Google Research Blog: VisualRank