ODU sets pace to better ID a face

NORFOLK

After the 2001 terrorist attacks, facial recognition software became one of the technologies to get fast-track federal and industry funding.

It was touted as a way to identify terrorists at airports and in crowds at large events such as the Super Bowl. It might even be used at the inauguration of President-elect Barack Obama.

But civil libertarians raised red flags over possible privacy invasions, and facial recognition, as developed so far, has proved unreliable.

Now, an ODU engineering professor wants to change the perception of the technology.

"We need 100 percent accuracy," said Vijayan Asari, an Old Dominion University professor of electrical and computer engineering. It's a lofty goal that could come out of the second-floor laboratory at ODU's Vision Lab, which Asari directs.

Asari and his graduate students are seeking to eradicate the concerns of civil libertarians, who complain that facial recognition technology will scoop up innocent bystanders and lead to a "1984"-like society where cameras record our every move.

That's not where Asari says his technology program is headed.

"We are not interested in recognizing every person in this country," he said this week at the Vision Lab, on Monarch Way.

Funded largely by the Department of Defense, Asari's facial recognition technology is aimed at battlefield and homeland security uses. If it proves reliable, it could be used in law enforcement and for corporate security as well, he said.

Current facial recognition software works by matching digital pictures taken from cameras situated in high-security areas with databases of terrorists or criminals.

The American Civil Liberties Union denounced it as a technology "with deep privacy implications that at minimum should be the subject of a thorough and wide-ranging debate before any use."

Since 9/11, Congress has spent millions each year on the technology but has failed to hold significant public hearings on the topic, despite notoriously poor results, according to congressional records.

Existing uses have been unreliable and have had questionable benefits. Virginia Beach used a facial recognition system - unrelated to Asari's project - for about four years until October 2005, when it was shut down because of performance and technological problems.

A similar system in Tampa Bay, Fla., was shut down for similar reasons.

Officials with the Electronic Privacy Information Center have expressed skepticism over claims of 100 percent accuracy.

John Verdi, staff counsel for EPIC, is unfamiliar with the program Asari is developing but said any similar program would need to undergo close scrutiny and peer review before it could be accepted as foolproof.

"The reality is, when you look at any of these technologies, I don't think we've seen one that's 100 percent accurate. That strikes me as an ambitious claim," Verdi said.

The technology being developed in Asari's lab goes well beyond matching photographs. Instead of focusing on, say, the nose and lips, the software divides the face into more than 1,000 images, creating a three-dimensional view. Asari said he hopes to be able to identify someone even if he or she is wearing sunglasses and a partial mask, or is hidden by shadows, or is running.

He has 24 graduate students in the lab. Some, but not all, work on facial recognition.

During a recent visit, one student demonstrated a robot, being developed for military use, that can drive itself. Another showed off a vessel recognition system - something that could have been useful to identify the rogue boat that blasted a hole in the guided missile destroyer Cole in 2000, Asari said.

But facial recognition remains the cornerstone of the lab.

As of now, Asari believes the technology is about 90 percent accurate.

But it's got to be 100 percent.

"Even 99 percent accuracy is unacceptable," he said.

Tim McGlone, (757) 446-2343, tim.mcglone@pilotonline.com