My colleague showed me a really fascinating image search by colour that indexes 10 million Flickr images and allows search based on colour.
For example here are the results for the colour blue:
And I can break down these results even more by adding the colour red:
The people who made this Flickr image search are "developers of advanced image recognition and visual search software" and i've also discovered another search engine they have called TinEye which searches for copies of an image you give it. Handy for copyright checking.
Image search is a fascinating subject and one that hasn't really grown up as much as text search. Things like Google Image Search are very much based on titles of images, related text, alt tags and link anchor text because of course search engine spiders traditionally only read text and can't 'view' images like a human can. STOP PRESS - Update: Google have just launched search by colour on Google Image Search.
Google Image Labeler is one method of getting round this problem - using humans to tell the computer what things are through the use of a simple game. Using humans to categorise images is a simple way to better search - Wikimedia Commons has a nice tidy pile of multimedia all categorised and named up as appropriate.
However I do like the idea of trying to search images in different ways - by using colours like the above example is one way - perhaps by the angles contained within the picture? Perhaps a search could be created that 'recognises' commonly seen shapes - such as buildings, cars, people, animals, hair, pornography!
Is structured data the future of search? (Mar 2010)
Don't just have alt tags on images, include height and width too (Nov 2007)
Give images rounded edges using Photoshop (Nov 2007)