Innovations: Finding familiar faces

New tech by a group of 'Garage Geeks' can do mass-scale facial recognition on low-quality images.

face recognition group 88 248 (photo credit: Courtesy)
face recognition group 88 248
(photo credit: Courtesy)
On the surface, it might seem like just another fun Facebook application that scans and organizes photos of you and your friends. But once you give a try, it's hard not to be amazed. This new facial recognition technology is an incredibly powerful tool with far-reaching implications. Founders Gil Hirsch, Yaniv Taigman, Eden Shochat and Moti Shniberg came up with the idea at a group called Garage Geeks. The group hosts events, meetings, workshops and special guests such as Sergey Brin, the cofounder of Google. After much brainstorming about how to build a technology that could successfully do mass scale facial recognition with low quality images, they spent the last year and a half developing the new technology from scratch. Their original plans were to start with existing technology in the security arena, but they quickly discovered that when it comes to everyday picture albums, the same parameters no longer apply. "As soon as you have a tilted head or an open mouth or you're wearing a pair of sunglasses, the existing security technology that works with full-frontal biometrics fails to recognize people," explains Hirsch. "We are the first ones to use a combination of technologies that focuses on recognizing people in everyday photos with great results." The goal is to maintain a 90 percent level of accuracy and the new technology won first place in a "Faces in the Wild" experiment conducted by the University of Massachusetts' Computer Vision Library. Surprisingly enough, even though it is still in alpha development and admittedly working out a few remaining bugs, the application found and correctly tagged pictures of me and friends of mine whose faces were partially hidden, blurred and dark with great accuracy. On another positive note, it's easy to install. With the click of a few buttons in your existing Facebook account, the photo finder begins to scan images of you and friends in your social network. It returns results relatively quickly - especially considering the thousands of photos it processes. This scalability, in fact, is another innovative element behind this new technology. And the numbers are gigantic. Nearly 1 billion images a month are uploaded to Facebook and right now is able to perform facial recognition on all of them using only a few machines. Since its launch, it has scanned 401,076,122 photos, identified more than 700,000 faces and logged 154, 815 tag confirmations. ALTHOUGH APPLE'S iphotos and Google's Picassa offer the same possibility to scan and label faces, they only include images in your personal collection.'s photo finder scans photos of you and your friends, including albums in your wider network where you and your friends are likely to appear. With its photo finder, you can also manually search for images of friends or browse for photos that were recently tagged. Other features include a "watch list" in which you can get notifications of new images of specific people as soon as they are uploaded. You can also sort through images by either date or accuracy and the application lets you know whether the images were already tagged in Facebook or found by the photo finder. Because it uses Facebook's existing platform, it's an automatically viral tool that is fun and engaging. I particularly like being able to see images of friends all in one place rather than having to go to their profile page to look at their albums. Some people are concerned about privacy with this kind of application, but photo finder maintains Facebook settings and does not store images on its own servers. In other words, if your images are hidden from certain friends, they cannot use the photo finder to retrieve your pictures. On the flip side of the privacy issue, this new technology can also be used to find images of yourself that others upload of you, which you can then either tag or remove so that others will not be able to view them. The photo finder uses crowd sourcing to help improve its accuracy. Once it scans and labels images in your account, it allows you to either accept or decline the tag. Therefore, if it confuses your grandmother and your mother, you can correct the mistake. It also tells you the percentage of accuracy beneath each image. THE INITIAL response to the new application has been overwhelmingly positive, but one of the biggest complaints is that images you tag in's photo finder are not automatically transferred to your Facebook account. According to Hirsch, they are working on an integration now that will give users this capability in the future. "Our goal is to help people find images that they would not otherwise see," Hirsch says. "We're offering people a fun and engaging way to share photos." The company is privately funded and although it is currently in stealth mode regarding future plans and monetization, Hirsch maintains that the Facebook application is only the beginning. "Facebook was a great place to start because it's a great platform to launch in for facial recognition of everyday photos, but there are many other applications for this technology on-line because it fills a great need," he says. "We're not stopping here." It's simple to install, engaging and provides a great, free service. As it's still in private alpha, you need to sign up for a trial on its Web site. Give it a try. It's impressive and addictive.