Colour chase
BY HOOKING on to computer networks like the Internet, the amount of information you can access through your computer is mindboggling. But this glut of information comes complete with a special, needle - in- a- haystack problem: the retrieval of relevant data, especially when the information is image-based. Now, researchers at the University of California, Berkeley, have developed a prototy pe of a digital library system called Chabot, that would help users ram through the binary mob and get the visual information they want (Science, Vol. 267, No 5203).
Unlike other systems which search with tags or keywords, Chabot . computes and stores the colours of the image itself. For instance, if the user wants to access all images which show a pink shirt, the user can search for a keyword, say, "shirts", and then add that the image has some pink. The system then lassoes all the images with shirts in them and combs them for pink.
In fact, it does even better. You can even scan pictures of pink shirts into the computer and then ask the system to took for similar images.
But even Chabot leaves room for improvement. For instance, the keywords themselves have to be cross-referenced with similar words so as to not miss relevant images which may be catalogued differently. In the above example, "shirts" should relate to "apparel", "clothing" or "garment". As of now, the system uses about 4,000 such semantic concepts; the developers hope to increase the tribe of concepts soon, although precisely how the user will cut through the babel is a problem yet unsolved. In any case, it is only a matter of time before retrieving information from giant networks will become as easy as picking your way through a library.