The FaceIt system by Visionics uses a security camera, local feature analysis algorithms, and a registered database to identify people in a crowd.
Events of the past several months have put the spotlight on security applications. Now photonic technology, in conjunction with target recognition algorithms, is helping facilitate rapid identification of faces in airports, train stations, and other public areas. "Face-recognition technology has the advantage in that it does not require active participation from people," says Prianka Chopra, analyst at Frost & Sullivan (San Jose, CA). Chopra estimates that the global face-recognition biometrics market was worth $6.6 million in 2000 and is growing at a CAGR of 65%.
Visionics Corp. (Jersey City, NJ) fields an identification system called FaceIt. "If you look at the human face, it is a reducible global pattern," says Joseph Atick, founder, chairman, and CEO of the company. He notes that the human face consists of features or "landmarks" such as cheekbones, eye sockets, and the bridge of the nose, which appear again and again among people. "What identifies us is how our features are spatially composed relative to each other," he says. Using a series of complex algorithms, FaceIt identifies these reducible components.
In most airports or places that require a security camera, not all of a person's face is visible. "But there is always a subset of about 20% of the total features," says Atick. "We found that through triangularization, we can reconstruct the face and locate all the other landmarks from this subset." The process yields a complete topographical map of a person's face described in 84 bytes of data (vectors). The software automatically compares vectors or prints from new faces with a database of known, registered face prints. A metric dictates how close these vectors have to be in order to achieve a match.
"Everyone's face print is unique," says Atick, "while the image of your face may not be. We can change our facial features through makeup, contact lenses, or weight, but we can't change the protrusion of our skull." With the Visionics system, a standard television camera captures images and sends them to the FaceIt converter, which continuously runs the 'find' and 'template creating' algorithms, converting every face that passes in front of the camera into a face print. The network then transfers the print to the matcher database for comparison. If a match is made, an operator in the central control room will see an image of the person matched against the database image and can notify security. The system can perform one million matches per second with a single processor.
David Watkins, president of Advance Biometric Images LLC (Tampa, FL), installed the Visionics FaceIt system for the Ybor City, FL, police department, placing 36 cameras in a 12-block area of the city's busy entertainment section. The system searches for more than 49,000 registered faces, including runaway teenagers, people who have skipped bail, or known felons. In a suburb of London, the crime rate fell by 65% when the city installed FaceIt. Watkins hopes for similar results in Ybor City. "I haven't seen any other face-recognition software take the technique to the level of accuracy that Visionics has," Watkins says.
However, there is still room for improvement. The main challenge of the system is noise. In order to classify a pattern, landmarks are essential. But these are difficult to identify if someone moves too fast in front of the camera, creating a blur. So the software must compensate. Variability of lighting is another concern. "If the light is bad, it may interfere with the algorithm's ability to find landmarks," says Atick. To compensate, the system scans regional information instead of looking for discrete points. "With this method, the effect of light is not as pronounced as if the system were relying on one or two pixels," says Atick.
FaceIt is fast becoming a fixture in airports around the United States. "It is a toola componentjust like scanning luggage," says Atick. "The beauty of it is that it works from a distance, in a crowd, without asking for participation. It buys a level of security without inconveniencing the public."