- The Washington Times - Thursday, March 21, 2002

The latest in post-September 11 security software never forgets a face. Face-recognition systems, which store images of faces to be compared with images gleaned from video cameras, give modern security experts new weapons in the war against terrorism.

The technology's recent rise is an obvious offshoot of last year's atrocities, though face-recognition software has existed for several years. The software also does more, potentially, than spot suspected terrorists in crowds.

Face-recognition systems are in use worldwide to help find missing children, crack pornography rings and provide businesses with employee-recognition systems.

The jury is still out, though, on how sophisticated each system can be and whether the technology works efficiently enough to merit its use in public spaces.

Face-recognition software typically involves a computer system with a bank of stored facial images, video images of people's portraits digitally transferred into mathematical data. One or more video cameras watch over an area in question, whether it is an airport check-in counter or a busy downtown commercial hub, analyzing all the faces that cross the camera's path. If an analyzed face bears enough data matching a particular face in the databank, the system alerts the user that a match has been made.

Michael Bonanno, an adjunct professor in security management at George Washington University, says much of the face-recognition software works by analyzing key constants in a person's face.

"Some things don't change the distance between your eyes, the width of your nose, the shadows of your face, the width of your eye," Mr. Bonanno says.

• • •

Face-recognition science falls under the broader term of biometrics, which also includes fingerprint and iris-pattern identification.

"With a fingerprint, you need to get 14 matches for it to be a possible match," Mr. Bonanno says. With face-recognition software, typically "several dozen matches" are needed before an alarm may be tripped.

All faces can be broken down by specific facial landmarks, or nodal points, aligned to the bone structure of the face.

Taken as a whole, the spaces between these nodal points give a specific identity for each person, distinct from any other, like a fingerprint. These nodal points, proponents of face-recognition software say, resist changes wrought by aging, skin tone and hairstyle.

The enrollment image, the original image in the system's database, should be as crisp as possible for maximum effect. A snapshot taken under duress won't be as valuable, or as accurate, as one taken under more ideal settings.

Mr. Bonanno says the recently completed Super Bowl in New Orleans included a degree of face-recognition software as part of security measures.

Typically, the software helps with security, transaction authentication and access control.

It does promise other uses, though.

England's national crime squad is creating a database of children's faces, culled from child-pornography rings, to help investigators match them with files of missing children. Viisage Technology, a Littleton, Mass., firm, earned a $1 million contract from an anonymous U.S. state to incorporate its face-recognition software into the state's driver's-license database.

Joseph Atick, chairman and co-founder of rival Visionics Corp. in Jersey City, N.J., says the technology has been in development for the past 15 years.

Mr. Atick says face-recognition software improved by steps during the 1990s, transforming it from a "cool toy" into software with real applications, particularly for businesses fretting over identity fraud.

Speed was a key stumbling block in earlier versions of its software.

Early Visionics software could match a scanned face to 10 to 20 faces in its memory per second.

After some fine-tuning, "we reached 1 million per second … that made it viable for large-scale programs," Mr. Atick says.

He dismisses criticism that a fake beard or other facial camouflage can trip up his system.

"It's studying the structure of your skull as it protrudes through your skin," he says. "The skull has about 80 landmarks visible from ear to ear. You don't need all of them to establish facial identity."

The system works best in a controlled environment, something many locations, such as airports chockablock with patrons, won't supply.

"If you have no control over imaging conditions … it still delivers some value if you're willing to deal with a high level of false alarms," he says.

Visionics equipment is in use, in pilot form, at Dallas-Fort Worth and Palm Beach international airports. Faces are scanned at various parts of the airports and fed into the software system. Several governmental agencies in the District also use Mr. Atick's company's software, but he declined to get specific.

• • •

Not everyone is convinced of face-recognition's accuracy, both now and down the short road.

Hany Farid, assistant professor of computer science at Dartmouth College in Hanover, N.H., says levels of accuracy are difficult to pinpoint.

Faces obscured by dim lighting, a large cap or a person's hunched frame can give the software fits, Mr. Farid says.

The main question is, "How often do you see the person you're looking for, and how often do you misclassify them," asks Mr. Farid, part of the college's Institute for Security Technology Studies. "That's the real concern. How often are you wrong?"

Mr. Farid sees a ceiling for how much improvement to expect.

"I don't believe you can take this really blurry image … and get this high resolution. I don't see that happening," he says. "These systems will never be foolproof."

Richard Chace, executive director of the Security Industry Association, says September's terrorist attacks brought a new level of attention to the field.

"The results of 9/11 really encouraged people to look at technology we previously wouldn't have looked at, for cost or social-stigmas reason," says Mr. Chace, whose District-based trade association represents more than 300 security companies worldwide.

He says airlines, nuclear facilities and pharmaceutical laboratories are among the businesses looking into face-recognition software.

"The technology is good technology, but you're talking about processing very complex algorithms. That can be extremely expensive," Mr. Chace says.

Current accuracy levels need to be improved, he adds, from current success rates around 45 percent to at least 60 percent before the technology will be embraced.

Concerns exist whether face recognition represents a tool against terror or another infringement on rights. Mr. Bonanno argues that a certain part of the debate is answered by the technology.

"The software is there as a tool … it doesn't constitute unlawful searches," he says. "In the public space, there is no expectation of privacy. It doesn't store the image of the individual it scans. That's what people are afraid of. If there's no match, it gets rid of it."

Either way, he says, face-recognition systems will affect the way many companies do business.

"How convenient would it be for an executive to be greeted by a sensor with 'Good morning, Mr. Jones,'" he says, as the door opens automatically. "That can be done through facial recogition. The cost right now is tremendous to implement something like that. That's down the line."

Copyright © 2019 The Washington Times, LLC. Click here for reprint permission.

The Washington Times Comment Policy

The Washington Times welcomes your comments on Spot.im, our third-party provider. Please read our Comment Policy before commenting.


Click to Read More and View Comments

Click to Hide