Biometrics Enters Third Dimension 

By Lakshmi Sandhana   |   Also by this reporter

02:00 AM Jan. 23, 2004 PT

A three-dimensional mug shot may soon be the only ID you'll ever need.

DuPont Authentication Systems and A4Vision, a company that sells facial-imaging products, have developed a biometric security device that generates in-depth, three-dimensional facial portraits similar to holograms and secure enough to be embedded in documents.

Using A4Vision's Enrollment Station, people can have their 3-D facial image embedded in a film called Izon and registered as digital data in a database in less than 10 seconds. The device outputs both a 3-D biometric template and a standard color image of the person.

The image in the biometric template carries enough detail to view a subject's head from ear to ear. The template can be affixed to cards or passports; once the image is embedded, users need only be scanned to see whether their facial characteristics match. The biometric data obtained is more comprehensive than 2-D imagery since it contains information along three axes instead of two.

"When I first got into biometrics over two years ago and looked at facial recognition, it was clear to me that in order for it to be truly effective, it was going to have to be 3-D," says C. Maxine Most, principal at Acuity Market Intelligence.

To capture the 3-D image, subjects stand still in front of a digital camera for three to 10 seconds while a projector beams an invisible coded light pattern onto their face. The camera captures a video image at the rate of 25 frames per second. A biometric snapshot is generated using reconstruction algorithms to calculate the difference between the initial coded light pattern and the final scanned pattern that registered the distortions caused by the person's facial geometry.

Sensitive enough to differentiate between the reflections caused by bone and soft tissue, the algorithms are able to accurately reconstruct a face.

Once photographed and enrolled in the database, a person approaching a camera or a security checkpoint apparently can be identified accurately in less than five seconds within a 6.5-foot range.

"I believe that the matching of 3-D images can probably be made more accurate than that of conventional face recognition using 2-D imagery," says Donald P. D'Amato, a biometrics expert at Mitretek Systems, a nonprofit research organization. "However, the set of 3-D and 2-D features that are chosen will be crucial to the level of accuracy achieved."

Right now the device is said to be accurate enough to distinguish between identical twins. Working with SRI International's twin registry, the company has tested the device with 36 twin sets, and it was able to distinguish one twin from the other.

However, David West, CEO of Geometrix, a company that also sells facial biometric products, does not think this is an exclusive development. "All 3-D systems can claim this capability. Twins become easier to distinguish as they age. No one has completed any studies that show the effective ability of these systems to differentiate at all ages."

Accuracy is a big concern. Identity theft appears to be the fastest-growing crime in America, with identity-related crimes projected to rob the global economy of $24 billion this year.

If not well-protected, biometrics may cause even more spectacular cases of ID theft, such as the gummy bear fiasco. Evans says 3-D facial identification is secure, however, because the facial image is only stored with the holder of the biometrics data.

"It takes the card or 3-D facial image, the holder and the database to match before security can be breached," says Grant Evans, CEO of A4Vision. "So just swiping a card won't allow me to use someone's credit card anymore, for example. Or just breaking into a database will not supply me with sufficient data to construct a 3-D facial image."

But is it spoof-proof?

"No biometric is totally spoof-proof, but we focus on hard tissues, which require extensive work to change," says Evans. "The issue here is: Can someone not be recognized? Yes ... but can someone fool the system into thinking it is someone else? No."

Because A4Vision's algorithms are tuned to measure hard tissue -- such as the span between eye sockets, the bridge of the nose, the span from jaw line to eye socket and temple to temple -- plastic surgery poses less of threat to the system's accuracy.

"Hard tissue is not often altered in plastic surgery," says Evans. "Eye sockets cannot be changed, or you will lose your vision. Not to minimize it; a face might be altered to be unrecognized on one axis, but not all axes. There would be enough data to indicate from 60 (percent) to 80 percent identification of a known terrorist, for example -- enough to raise a flag."

No terrorists, though, are going to willingly line up to have their faces digitized for eternity.

Solutions put forth involve setting up a modus operandi to capture potential suspects' faces covertly or making the use of such systems mandatory with all forms of ID -- driver's licenses, national IDs and the like. A4Vision has been working for the past two years on conversion programs aimed at taking the huge database of existing 2-D images available on IDs like licenses and converting them into 3-D files. The company says the information extracted can be used to provide biometric verification. While it's been tough to achieve 90 percent accuracy consistently, the company is getting closer.

It's not going to be easy, though. "It's important to realize that 3-D face recognition is a completely new biometric," says Andy Adler, an assistant professor at the University of Ottawa. "That means that it will take 10 years, minimum, before the community learns when it works, and when it doesn't."

"The biggest obstacle for the adoption of any 3-D facial-recognition technology is that there are no existing databases -- unlike those of fingerprints or 2-D mug photos," adds Acuity's Most. "These 3-D databases must all be built from scratch."

Deploying dual biometrics is most likely to emerge as a standard, utilizing a combination of biometrics to provide unmatched security.

A4Vision has been engaged by the federal government to develop 3-D biometrics for homeland security applications, and the enrollment station is currently being tested by the US-Visit program.