Multiple Biometric Identification

IEEE Fellow set up a suite of algorithms that generate digital patterns from images of a person’s face, fingerprints, and even the layout of veins in the palms of the hands

6 August 2009

Mistaken identification happens all the time. But despite many years of research aimed at creating pattern-recognition systems that distinguish people by their fingerprints, the shape of their faces, or the arrangement of their facial features, none of these systems is accurate enough to allow police to make judgments about guilt or innocence. Still, technologists haven’t given up.

One way to build a better bionic mousetrap is by combining systems that specialize in different body parts or arrangements of parts. That way, the weaknesses of each approach tend to cancel each other out while the strengths add up. The unified system should be less likely than its parts to collar an innocent person or free a guilty one.

Rangachar Kasturi, a professor of computer science and engineering at the University of South Florida, in Tampa, described such an effort at an IEEE media event held last March in New York City. Kasturi, an IEEE Fellow, and his team are setting up a suite of algorithms that generate digital patterns from images of a person’s face, fingerprints, and even the layout of veins in the palms of the hands. The system could only mistake one person for another if all of the digital files representing bodily characteristics were exactly the same in each person.

One biometric identification hurdle that the Florida team particularly wants to clear has to do with identifying people by the way they walk. Such a capability would work at a greater distance and under worse lighting conditions than a facial-recognition system could tolerate. Kasturi says it would be a perfect complement to the other characteristics that can now be tracked. The main outstanding question, he says, is how to decide which aspects of a person’s gait should be captured and which can be ignored.

Kasturi’s team is also trying to improve the way their system analyzes digital image files. Such an improvement would pay off even in a system that examined only one characteristic. For example, his lab has written programs that analyze images captured by 1- or 2-megapixel cameras on cell phones. It works with the feature that allows many cell phones to store images of contacts next to their names and phone numbers. Say you’ve met someone before and feel embarrassed about having forgotten his or her name; you simply snap another picture of the person when you meet again, and Katsuri’s software will search through all the images associated with your contacts until it finds a match. Face recognition works well enough with a cell phone because most handset makers limit the number of contacts to 500—a small enough set to ensure a low incidence of false positives.

MORE THAN WHODUNNIT Kasturi’s pattern-matching methods can also help to improve security. A computer programmed to tell whether the glide in one person’s stride is the same as another’s will handle a job for which humans are ill equipped: round-the-clock monitoring of surveillance cameras. Psychologists have demonstrated that people who must monitor four video screens typically miss 20 percent of whatever it is that they’re supposed to notice; those who monitor nine screens miss 50 percent. And that’s when they’re well rested; people do even worse when they’re tired. Computers, of course, never get tired.

Kasturi’s team is using the same pattern-matching technology to develop devices that translate American Sign Language into speech or text. They will work the same way the face-matching program does, but the database will be populated with ASL’s hand gestures and the words or phrases they represent.

In his talk, Kasturi mentioned that users of YouTube will be able to call up clips based on descriptions of the actions they are looking for—jumping over a speeding car or swinging at a baseball and accidentally throwing the bat, for example—instead of having to guess at the keywords the video’s creator used to identify the uploaded video.

Kasturi says the U.S. military has also expressed interest. His team has developed software that detects power lines captured on video, then displays them as bright yellow lines on monitors so that pilots can steer around them. That’s important, because the U.S. Army loses more helicopters and remotely controlled aircraft to collisions with power lines than to enemy fire. Kasturi and his colleagues are also working on a submersible ocean-surveillance device that will detect ships by their sound, pop up to the surface, take pictures, and send the images via satellite to the Coast Guard.

Kasturi admits that there is still much work to be done before the various technical tracks dovetail into an ability to identify criminals, exonerate the innocent, spot unattended luggage at airports, or alert authorities when someone authorized to be in an area begins doing something that person isn’t authorized to do.

Watch Kasturi’s talk on IEEE.tv.

Learn More