At a Web conference in Beijing, Google scientists presented a paper on software that can search and rank digital images much the way the company's search engine does for text, reports Times Online.
So far, image search has largely been done based on captions and other text around it, a method TechCrunch refers to as "trying to describe the color green when you can only see in red."
Rather than identifying a cow by an outline of a cow, the Times story says, the program would discern a pattern or "visual theme" to images and would be able to link, for instance, a shot of McDonald's golden arches as well as an interior restaurant shot of a McDonald's.
The search would allow shoppers, for example, to focus on a particular style of brown shoes or buckle on the Web and find similar products from competing merchants, reports The New York Times. With all the images out there to be cataloged, however, there's no word when such a search will be offered to the public.