People Search Digest

Did you know that image searches are basically just glorified text searches?

When a search engine tries to find an image for you, it’s not analyzing image, video, or audio content; the only data it can use is the text associated with image files. As you must expect, this presents an incomplete picture — no pun intended. In the past, Google has tried to rectify this problem with its Image Labeler game, which asks actual users to assign proper tags to image content. A regular Tom Sawyer, that Google. But it still comes down to trying to describe images with text, an imperfect arrangement. Ezra Pound agonized for years over the futility of using words to represent imagery and was never quite satisfied, so Google probably realized that they’d better figure out another way to do it.

Two Google scientists recently presented a paper describing their attempt to use computers to analyze the actual images themselves, rather than simply the associated text. The new technology hopes to employ image recognition software to determine what “stuff” is in the searchable image. Once that’s achieved, Google will compare similarities between images, using the popular images as starting points and determining rank by comparing them to subsequent images. This way, the content of the image itself will be the determining factor in an image’s ranking, rather than the associated text.

Preliminary testing of this new engine has shown significant improvements in user satisfaction, so it seems Google’s on the right track. Provided they continue advancing, this could prove to be the first step toward the analysis of rich content by computers and a new age of online search in which humans no longer do the work.