Image search has left something to be desired for the very good reason that it’s really tough for a machine to figure out the content of an image.
Search engines have traditionally looked for metadata surrounding a given image when determining the relevance of a graphic file to a search query. That data has include keywords within a graphic’s file name, the alt text associated with the HTML image code, and the text surrounding the graphic such as captions. Everything, that is, but the actual content of the graphic.
In an effort to boost their ad revenue from their image search function, Google is developing technology that would allow them to scroll the pixels of images in order to recognize scenes and faces.
Beet.TV interviewed R.J. Pittman, Google’s Director of Product Management for Consumer Search Properties, about the technology:
He sure does say “properties” a lot, doesn’t he?