So Google Image Search is great when you can express what you’re looking for in text, like "cat on roof" or "map of Kansas." But what if you want an image similar to one you already have? Wouldn’t it be better to search using the image?
For content producers, it might also be important to search the Internet for images identical to images they own to protect against copyright infringement. Performing that kind of search using text would be difficult at best, and impossible to automate.
This is not really my research area, but a few years ago I needed to write a research paper for a class and hit on the idea of using fingerprint recognition algorithms for image search. Thus the incredibly wordy title of (deep breath):
Self-important titling out of the way, it was quite an interesting technology evaluation to perform. Of note, several of the companies I looked at also sell facial recognition software, and some may even produce large-scale image search systems now; I haven't kept up with the market.
Most interesting of all, though, is the productization of image search: a site called TinEye takes an image as input and returns a list of similar images, along with their location on the Internet. This, of course, is a long tradition in computing: yesterday's research project is today's hot product and tomorrow's old hat. Fun to see it in action, though.
Update: Google image search using images was just announced today at Google's 'Inside Search' event. Interesting that this space is starting to get populated now!