The facial-recognition software used by the FBI can lift shadows, sharpen blurry pictures and create three-dimensional images from flat ones, Albers said. It also analyzes each face to develop a unique “template” based on the shape, skin texture and distances among features.
Still, facial-recognition software remains imperfect. In most cases, the human brain is more adept at identifying individual faces, experts say. Computers are most valuable when seeking to compare an image with many possible matches, such as in a database.
The FBI, which declined to comment about the use of such technology in the Boston case, ultimately turned to the public in its quest to identify two key bombing suspects, even though it later turned out that one of them had a driver’s license photo on file in Massachusetts — one of at least 30 states that use facial-recognition software to analyze its database of motorists.
Though the software is used mainly to detect fraud when people get more than one driver’s license, the rate of false positives remains high. Claims about precision often are overblown, say those who study the technology.
“The public has grown up in a world of ‘CSI Miami,’ ” said Marc Rotenberg, executive director of the Electronic Privacy Information Center. “The real world is much more complicated.”
In one case of mistaken identity in 2011, covered at the time by the Boston Globe, a suburban Boston man received a letter from the state Registry of Motor Vehicles telling him that his license had been suspended.
Only after he hired an attorney and missed more than a week from his job as a truck driver did officials explain that a facial-recognition program had concluded that he and the holder of another license were the same man. The error was corrected and the man’s license reinstated after a contentious hearing and the filing of new identity documents.
“It was a really nasty, unpleasant experience all around,” said the man’s attorney, William Spallina.
Whatever limitations exist in the software are likely to be surmounted in the next several years. Engineers are working to improve the clarity of images and the ability of software to recognize abnormal behavior before an incident occurs. A future generation of analytical software might, for example, identify a man putting down a backpack at a busy event and immediately alert police.
“There are certain situations when you give up some of your privacy, and we’ll be seeing that more and more,” said Albers. “The old idea of privacy . . . certainly is changing.”