 |
 |
Face Recognition Needs More Study
By Robin Herron
Shortly after the Sept. 11 terrorist attacks, television viewers saw videotapes
of the hijackers taken by airport security cameras. Among the many questions raised
in hindsight was, Why couldn’t face recognition technology have been used
to identify the terrorists who were listed on the FBI’s Most Wanted List
before they boarded the planes?
That question also has been posed to Harry Wechsler, professor of computer
science and director of the Center for Distributed and Intelligent Computation,
which includes the Biometrics and Forensics Lab. His research group has been studying
face recognition technology since 1993, when it first participated in a government-sponsored
Face Recognition Technology (FERET) Program. The program’s goal was to develop
face recognition algorithms and a large and representative facial database for
testing the algorithms. Wechsler, together with his associates, developed the
FERET database, which has become the standard used to test face recognition algorithms
around the world.
However, the FERET Program showed that the technology’s reliability decreased
over time. Because the subjects and settings changed slightly as time passes,
newer images did not exactly match the images in the database. The technology
recognized only about 50 percent of the subjects 6 to 12 months after the original
image was taken.
As a result of this research, Wechsler has strong doubts about the technology’s
usefulness by itself in identifying terrorists—or anyone else—in uncontrolled
environments. “The technology is in its infancy,” he says. Wechsler
recently discussed the technology’s pros and cons on WJHU, a National Public
Radio affiliate in Baltimore, Md. Other guests on the show were the president
of a leading manufacturer of face recognition equipment and a privacy rights advocate.
During the show, Wechsler pointed out a number of problems with the technology,
including the fact that targeted persons must be pre-enrolled in a database and
obtaining clear photos of suspected terrorists is problematic. He also noted that
changing one’s facial appearance or hair style or color is easy, necessitating
continual updating of the database. Finally, for the technology to be successful,
the environment must be controlled, with proper lighting, no distractions, and
no occlusion.
That’s not to say the technology can’t be improved. Wechsler and
his student Chengjun Liu, now an assistant professor at the New Jersey Institute
of Technology, have developed new and successful face recognition algorithms and
have applied for a patent. In addition, Wechsler is using video processing to
capture the whole human body and its behavior, providing more data such as upper
body and arm movements. “If I see someone at the end of the hall, I may not
be able to recognize her face,” he says, “but I might be able to recognize
her from the way she walks. The more clues there are, the easier it will be to
discriminate.
“It’s important to understand that [face recognition technology]
alone will not solve everything,” Wechsler says. Old-fashioned fingerprinting
is still an effective way to identify people, he points out, and new technologies
are waiting to be discovered. “This is our mission here,” he says.
For more information on this research, visit the center’s web site at
cs.gmu.edu/~ wechsler/DIC_Center.
|
