Real-time face recognition comes to your iPhone camera

Tagging friends after you snap a photo of them and posting it to Facebook is so last week. A new smartphone application allows you to point an iPhone camera at a friend and tag that person before you even hit the shutter button. Called "Klik," the iPhone app automatically displays your friends' names in real time when they appear in view of your iPhone's camera. After Klik detects a face, it instantly connects to your Facebook account and scans your friends' photos to identify the person in view. It also scans your iPhone for photos you've tagged on your phone. When a user snaps the photo, the subject is automatically tagged, and the photo can either be stored on the device or uploaded to a social network. The app was released on March 7 by Internet facial recognition service provider Face.com. The Israeli startup improved upon its "Photo Tagger" software, which finds friends' faces in photos and automatically suggests nametags for them -- a solution that Facebook adopted in its Photo Tag Suggest feature in late 2010. With Klik, facial recognition can now be done in real time. Face.com calls the Klik app "magical," but the software is by no means perfect. In field tests, it was only able to identify about a quarter of test subjects, even though they were connected to me on Facebook. In many cases Klik identified those people incorrectly. It had an easier time identifying photos, but that's kind of missing the point. Face.com admitted to "a few bugs," which it hopes to have sorted out for this week's South by Southwest conference in Austin, Texas. A spokesman for the company said that users can teach the app to recognize people by tagging them manually. He said some people are also using that feature to create private collections of tagged photos of people that are not connected on Facebook. Klik has a certain level of privacy built in, as the app will only identify people a user has previously tagged or people in a user's social network who have already agreed to connect and share with the software's owner -- presumably people the user already knows. Still, real-time facial recognition represents a direction of that is worrisome to privacy advocates. "Currently applications such as Klik tell you not to worry because it's through consent and just with your friends,' said Alessandro Acquisti, professor of IT and public policy at Carnegie Mellon University's Heinz College. "But in the long-run, there will be no technological barriers that would prevent something that could do this not just with your friends but anyone out in the wild." Today's technology is not quite robust enough to snap a photo of someone on the street and instantly know who they are. Computer processors aren't fast enough to scan across billions of images in real time to match an offline face to an online photograph. To match two photos of unconnected people in the United States in real time would take four hours, according to Acquisti. There are other hurdles as well, including the fact that computers have difficulty identifying faces in low light or from far away distances. Current face recognition technology also offers up too many false positives, evidenced by our field tests of Klik. Acquisti said those technological barriers will be overcome within the next 10 to 15 years. As processing speeds improve, face recognition experts believe apps like Klik will someday be capable of identifying most people in real time -- regardless of whether they're connected to you or willing participants. "In my opinion, this is one of the most problematic aspects of the trajectory we are clearly on," Acquisti said. To top of page