Page 291

An article on NewScientist.com describes new software being developed to help bridge the gap between verbal and visual searches. Pretty cool. Until now, most engines like Google have relied on tags attached to pictures, or the page in which photographs and other art are embedded, to produce image results. This new prototype begins with manually entered tags, then takes the next step and attempts to notice what researchers call ‘blobs’–similar visual patterns that tend to recur for a given tag–to develop semantic categories. Once the software has established a relationship between keywords and certain visual elements, it assumes that other, untagged pictures with those visual elements might be tagged the same way. Ultimately, this leads to verbal semantic relationships drawn from visual ones–a pretty neat idea, and one sure to lead to tons of fascinatingly flawed results. I can’t wait. I’d love to see a prototype in action.