Oct. 19--When a university expert in "biometric" identification decided to match wits with a CIA master of disguise, old-fashioned guile won out over modern technology.
Rama Chellappa, an engineering professor from the University of Maryland, didn't come out well in the scientist-vs-spy competition. His team tried to build a computer system that would identify the wily agent from a series of published pictures.
"He had false teeth he would use, all sorts of things," Chellappa said. "We only got 20 percent (correct matches.)"
Chellappa, director of Maryland's Center for Automation Research, was the keynote technology speaker at a conference at the University at Buffalo Tuesday that drew experts in ID technology from around the world. Biometrics is the science of determining identity through physical characteristics, like fingerprints or retinal patterns.
About 50 scientists attended the IEEE AutoID 2005 workshop on Automatic Identification and Advanced Technologies, a hot area of research in the post-9/11 era.
"This is a high-caliber biometrics group; I think it says a lot for the center that's been established at Buffalo," said Russell W. Bessette, executive director of the state Office of Science, Technology and Academic Research, who addressed the group Monday.
UB's Center for Unified Biometrics and Sensors was a co-sponsor of the two-day conference. The center brings scientists from different disciplines to work on ID technologies, director Venu Govindaraju said.
"Our focus is not only security, we're also looking at how biometrics should be able to take over the smart cards you carry, the PIN numbers we carry," he said.
The Buffalo area is also home to the biometric ID company Ultra-Scan Corp., which develops ultrasonic fingerprinting technology. The Amherst-based company on Tuesday announced it won a Pentagon research contract to develop a high-accuracy ID system combining fingerprint, retina scan, face recognition and handwriting technologies.
The signature on your check or credit card slip is one common form of biometric identification. Now, computerized biometric scanners are beginning to show up in building security systems and cash machines.
But while commercial ID systems proliferate, biometrics for advanced homeland security applications are proving more challenging, researchers said.
"Many of the systems on the market have a high false-positive rate, which could be misleading or even dangerous," Bessette said.
Surveillance cameras are everywhere, but computer systems that can use them to identify individuals in an airport or subway platform remain off in the future, Chellappa said.
Face recognition technology has unique drawbacks -- as the experiment with the disguise expert showed. "Faces smile; faces age," Chellappa said during his address. Even people not trying to evade identification pose challenges for computers, not to mention those with something to hide. Bank robbers have found that a baseball cap and sunglasses can help conceal their identity from conventional security camera users.
However, computer biometrics have yielded useful techniques short of matching identity, Chellappa said. For example, some automated systems are good at determining if someone is loitering or has dropped a package and walked away, he said, while gait analysis may indicate someone in a crowd is carrying a concealed package.
"Face recognition can be as good as iris recognition, but we have to prove that," he said.