Identity, Age And Emotion: Facial Recognition Garners Abundant Promotion

Facial recognition software developer Face.com has been in the news quite a lot in recent months. Back in late January, for example, the company launched a free iOS-based app called KLIK which "uses face recognition to let you quickly tag your friends in real-time. Fire up KLIK and watch as your friends' names instantly appear next to their faces before or after you snap a photo." My suspicion, although I haven't spoken to company officials to confirm, is that KLIK serves the same conceptual purpose for Face.com that GOOG-411 did (and Google Voice, formerly GrandCentral, may still do) for Google...providing the company with abundant data samples that it can use to improve its algorithms.

To wit and more recently, at the end of March, the company unveiled the ability to discern a photo subject's age. This claimed capability is, of course, an enhancement to the more crude child-versus-adult discernment that I wrote about back in early January. Face.com's supplied REST API has already been used by at least one developer in creating an iOS-based app called Age Meter. And the company's API, by the way, can now be leveraged to scan up to 5,000 images per hour (120,000 per day) at no charge. The company reported in mid-February that more than 10,000 developers are now using the API, initially launched in May 2010.

That's an impressive statistic. Equally impressive is what one of those developers has more recently done with the API. Dale Lane hooked a webcam to a media center computer running Linux, placed the camera underneath his television and pointed at him, and leveraged Face.com's algorithms to discern his emotion when watching television programs, playing Xbox-served games (note that the webcam was not a Kinect), etc. The webcam was only VGA (640x480 pixels) in resolution, and the script only grabbed an image every ~15 seconds. Still, the results are impressive. And in true open-course fashion, Dale even provides the source code on his web page.