The Future of Facial Recognition: Man vs. Machine?


photo: Andy Roberts, creative commons license

DCI Mick Neville of London’s Metropolitan Police Service runs a unit of people with superior facial-recognition capacity. He believes that image recognition will turn out to be the third revolution in forensic science, after fingerprint and DNA analyses. (This is part 2 of a 2-part story. Read part 1 here.)

Currently, the Met solves about 2000 cases a year based on fingerprints, another 2000 using DNA analysis, and 2500 with imagery recognition, at a tenth the cost of the other two techniques, he says. Writers of crime fiction have a lot to work with here.

Can’t Computers Do It?

Can computers eventually take over this job? People in the super-recognizer community say no. Part of the reason is the sub-par environments in which many closed circuit television (CCTV) images are captured. Says Patrick Radden Keefe in The New Yorker, “After the 2011 London riots, the Met gathered two hundred thousand hours of CCTV footage. Computer facial-recognition systems identified one rioter.” Gary Collins, one of the Met’s super-recognizers, identified 190.

Of course, computers are becoming more skilled all the time. Facebook’s recognition program, is touted as one of the best. Unlike CCTV, it mostly has well-lit, good-quality images to work with. It has a further advantage because it can narrow the universe of possible matches to the friends, family, and friends-of-friends of specific users. Yet even FB’s algorithm consistently identifies the wife of a friend of mine as me. When I look at her picture, I don’t see it, but Facebook does.

Computers definitely have some role, though, and the Met combines machine and human expertise. It uses a specially created computer program to narrow the number of images by broad demographics and type of crime, for instance, then lets the human super-recognizers make the match.

And, if facial recognition software is prone to error, Keefe says, logo-recognition algorithms work well. “It turns out that many criminals not only commit the same crimes again and again; they do so wearing the same outfits,” he says. That shirt with the six-inch polo player stitched across the left chest? Dead giveaway.

Where Next?

As super-recognizer approaches migrate to the United States (as they have already to St. Petersburg, Florida), authorities will need safeguards against false identifications. In the U.K., a case is never made against someone based solely on facial recognition evidence.

No one wants a repeat of the situation that occurred after the Boston Marathon bombing when the F.B.I. crowd-sourced the identification process, and innocent people were fingered. In these hysterical times, that could be deadly for false suspects (another plot wrinkle for us crime-writers). In the recent New York City/New Jersey bombings, a fingerprint had given them a specific name.

Hiring people for sensitive security positions at airports and nuclear power plants perhaps shouldn’t rely on the assumption that everyone is more or less the same in facial recognition skill, just as we don’t assume everyone is just the same in other job-skill domains. We test for those.

Because millions of Americans have little or no ability to recognize faces (see earlier post), researcher Richard Russell believes “it is statistically inevitable that some passport officers at American airports are face blind—and that quite a number are significantly impaired.” Why not make sure people in such sensitive positions are especially suited for these sensitive jobs?

Further Reading

Have a scientific bent? Here’s the research paper that started it all: “Super-recognizers: People with extraordinary face recognition ability,” by Richard Russell, Brad Duchaine, and Ken Nakayama, published in 2009 in the Psychonomic Bulletin & Review.