Brown Professor: Face-recognition when you shop, check in at the airport, or walk city streets

By G. Wayne Miller~Projo

It won’t be long, Michael J. Tarr predicts, before machines will recognize your face when you shop, check in at the airport, or simply walk the city streets. Your dishwasher or microwave may be programmed to say hello after confirming it’s you.

“It’s going to be ubiquitous,” he says.

Tarr, 47, is codirector of the Center for Vision Research at Brown University. His career-long work in understanding how the brain processes what the eyes see is part of a growing body of research that other scientists are using to perfect technology that can instantly identify a face.

The science of face recognition already is good.

To demonstrate, Tarr goes to his computer in his office in the Metcalf Research Laboratory, at the corner of Waterman and Thayer Streets. The building is home to Brown’s Brain Science Institute, which brings together specialists in neuroscience, psychiatry, psychology, computer science, engineering and six related disciplines.

Tarr logs into his account at Google’s Picasa site (http://picasa.google.com), where members can upload, store, edit, organize and share personal photographs. Many firms offer a similar service. What differentiates Picasa is its ability to recognize faces.

“Here’s Ben, my oldest son,” Tarr says. We see a shot of a smiling young boy.

When he opened his account, Picasa prompted Tarr to label the faces, which the computer had never seen, in a few of the photographs he initially uploaded. He typed in the names of his sons, Ben and Sam, and his wife, Laurie. The computer then asked if he wanted to find more photos with members of his family. Tarr did. In an instant, Picasa screened the rest of the photos –– and automatically identified his sons and wife, based only on what they look like. Tarr displays the many photos now in his account. Most of the people in them have been labeled –– by the computer.

The face-recognition software that Picasa uses, however, is not infallible. It could not label a small number of photos in the Tarr family album. Tarr shows them.

“You can see these are tricky pictures: weird profiles, kind of funny angles,” he says. One that stumped Picasa is an out-of-focus shot of a boy partially obscured by a toy. It’s Ben, the professor says.

“I can do it,” Tarr says. “I know his face well. That shows you why humans are still better than computers.”

Not for long. Tarr estimates that Picasa can correctly identify faces 80 to 90 percent of the time now, and as technology improves, the hit rate will climb. “Five years from now,” Tarr says, “it’s going to be high 90s.”

Already, programs not readily available to the public have been developed that achieve higher success rates than Picasa and a similar program offered to Facebook subscribers; they have military and law-enforcement use today. In the future, the average person will encounter face-recognition applications in daily life, Tarr says.

Take, for example, targeted advertising, which today is based primarily on non-image data.

“Imagine if they can recognize if you’re a male or a female as you’re walking through a store.” Or could tell if you’re Caucasian, Hispanic, Asian or African-American. The computer would instantly make the identification and trigger ads on wall-mounted screens that would appeal to your supposed tastes. Or send you ads or coupons to your iPhone or personal computer.

Workplace and home will also be affected, Tarr says.

“Someone clever is going to put [face recognition] in your appliances in your house, so it can recognize whether it’s a kid or the mom or the dad. Or imagine you have shared work stations at work and you sit down and it immediately knows which one of you it is.”

With traffic and surveillance cameras becoming fixtures of the urban environment, it’s no leap to envision the day — sooner rather than later –– when the police and other law enforcement agencies will use what is also known as “computer vision” to find and track suspects.

“Imagine all the security cameras in New York City,” Tarr says. “Imagine if they enable them with face recognition.” Now imagine police had a photo of the Brown scientist’s face in their database –– which they already could have, from Google or other Internet sites. “They actually could poll all those cameras and find out where Mike Tarr is,” he says.

Does this technology he’s had a hand in creating disturb him?

“It’s like anything: it depends on how you use it. If it’s pumping ads at you, that’s probably a downside. If it’s being used for what I would say is ‘overly secure monitoring’ –– Big Brother –– then it becomes a bad thing. If it’s used to solve crimes or it’s used to make things easier for you to interact with, then it’s a good thing.”

For those so minded, Tarr notes, foiling a computer would not be difficult; the human brain, he says, is easily tricked by disguises.

“I imagine a big system would be bad too,” he says. “If you suddenly put on a wig and a moustache and some dark sunglasses, it may completely fail. It couldn’t find you.”

At least not in the immediate future.

Tarr leaves Brown at the end of this summer to assume a position at Carnegie Mellon University in Pittsburgh, where he grew up. His research is described at http://titan.cog.brown.edu:8080/TarrLab/

gwmiller@projo.com