Does Face Recognition Have a Future?Privacy protections and reliability are the gatekeepers of controversial biometric technologyOf all the biometric technologies currently in use, face recognition is arguably the most controversial. If for no other reason, the interest in integrating a range of biometric applications with other authentication technologies such as smart cards and digital signatures means that face recognition will earn your attention. Challenges: Privacy and AccuracyAlthough gambling casinos have used biometrics extensively — scanning the room for known bad guys — applications elsewhere are less common. Borders Group Inc.'s bookstore has put its test program on hold, and privacy advocates are still fuming after the 2001 Super Bowl experience in Tampa, Fla. Cameras scanned the faces of people in the stands and matched them with pictures in a large database, leading Privacy International to award Tampa its 2001 Big Brother award (in the tradition of Oscars and Tonies, the statues are called "Orwells"). Since Sept. 11th, however, the public's mood has become disposed to finding a more lenient balance between privacy and security — a result that isn't altogether pleasing to civil liberty watchdogs. Be assured that smart people — from corporate leaders to the Department of Justice — are trying to determine how to negotiate the new environment. But the show won't go far anyway if face recognition technology can't deliver the goods. The major challenge, of course, is reliability. Ambient lighting, camera distance, and other factors can stymie accuracy. Facial expressions and disguises, say some critics, let too many false negatives get through the net. False positives, in addition, can wreak havoc if the rate is at 20 percent, and you scan a million people walking through a dozen airports around the country. Lining up 200,000 people at LAX, JFK, and other airports while humans check and filter the bunch for the needle in the haystack is a dog that won't hunt. Maybe even 1 percent is intolerable. So it is no accident that in the face of these challenges a planned test of rival "face grabber" software products, Visionics Corp.'s FaceIt and Viisage Inc.'s FaceFinder, at Boston's Logan International Airport this year has many people sitting up. Enabling TechnologyImprovements in face recognition technology will come not only from succeeding generations of software used for capturing images and comparing templates, however, but also from the hardware used inside the camera, such as new chips for video processing. For example, Equator Technologies Inc.'s image processor (in full disclosure, I am an advisor to the company) can enable enhanced image resolution, image sharpening, full-screen lighting normalization, and even facial rotation in a way that is better than previous methods. Designed as an image processor from the beginning, it can process video at four times the speed of a general purpose Pentium. This speed, in turn, makes the complex algorithms for accurate facial recognition a more realistic solution. As the increase in speed is accompanied by a decrease in chip power and cost (35 watts reduced to 2.5 watts; $350 reduced to $35), face recognition technology begins to morph into something far more practical than what exists today. John Setel O'Donnell, CTO at Equator, said, "New semiconductor technologies will power the growth of face recognition from a relatively small number of cameras to hundreds of thousands of cameras. The rising device count makes possible supercomputers on a chip that deliver the enormous processing power needed for realtime face recognition, small enough and cheap enough to go in every camera." Of course, some people will never expect inaccuracy to get much below a 0.5 percent to 1 percent. Among those same people, nonetheless, will also be those who'll promote face recognition for its deterrent value — if you're a criminal, would you really be comfortable with odds of 20 to 1, or even 100 to 1, of being caught? Alternatives and the Big PictureMost experts intuitively seek identification systems that employ several modalities, each having some combination of advantages and disadvantages. Iris scans are the most accurate today but are also the most expensive. Fingerprinting is already an established technology with law enforcement but not available for segments of the population whose fingerprints are worn, damaged, or inaccessible. Yet, the intriguing element of face recognition — and video surveillance more generally — is that unlike the other techniques, it is nonintrusive. You don't have to cooperate to have your face scanned in a crowd or a theft in the parking lot recorded. The job can be done quietly, even covertly. The implication for privacy protection is enormous, especially in the realm of nonsecurity applications. Imagine the day when a retailer wants to monitor a customer's behavior in the store with a camera loaded with face recognition and ties the output in real time to a sophisticated customer information system. One-to-one marketers will have a whole new ballgame to consider. As will privacy advocates. And you, too. Nicholas Imparato [imparato@hoover.stanford.edu] is research fellow at the Hoover Institution, a professor of management and marketing at the University of San Francisco, editor of Public Policy and the Internet (Hoover Press, 2000), and a director or advisory board member of several firms in the IT sector. RESOURCESEquator Technologies Inc.: www.equator.com Privacy International's Big Brother awards: www.privacyinternational.org/bigbrother Viisage Inc.'s FaceFinder: www.viisage.com/facefinder.htm Visionics Corp.'s FaceIt: www.visionics.com/faceit |
 |
Most Popular This Week
IE Weekly Newsletter
Subscribe to the newsletter
| |
|
|