Science World

Gotcha!

You can easily spot a familiar face in a crowd. But what if your mission was to pick out the face of someone you've never seen? Enter facial-recognition technology

March 7, 2003
By Libby Tucker

Ever searched for a friend in the middle of a mall or football game? It may seem like a miracle when you finally spot him or her, but have you stopped to think about how you managed to single out the one face you were looking for amid a sea of strangers?

Humans possess an extraordinary visual system capable of learning and recognizing thousands of faces -- even someone they don't know and have never seen: "If you want to train someone to recognize all 155 known terrorists on the U.S. government's list by their face alone, that's not a problem," says Bob Schmitt at Viisage, an identification-technology company in Massachusetts.

So in this age of heightened security from the potential threat of terrorism, you might think the government would train sharp-eyed special agents to comb airports or sports stadiums for suspicious characters. But, explains Schmitt, it would take thousands of trained officers to scan every face in a crowd for a person who may not even be there. Which is why the race is on for security companies to create revolutionary facial-recognition technology -- which uses a video camera and computer to distinguish a person based on his or her unique "faceprint." "It's the only technology that can identify someone who's merely walking down the street," says Trevor Prout of International Biometric Group.

Facial recognition isn't nearly as precise as a fingerprint or iris (colored area of the eye) scan. So scientists are probing the biology behind the human eye with a single-minded goal: to create artificial-vision technology that surpasses human ability and zeros in on possible terrorists, criminals, or missing persons faster than you can blink.

EYE-D

While every individual face may be unique, all faces share common features -- eyes, nose, mouth, and ears -- in roughly the same place. That's why the brain has evolved to detect miniscule differences between individuals, such as the particular shape of the eyes and mouth, explains computer scientist Gary Cottrell of the University of California at San Diego. We learn to distinguish one another's features through experience -- starting with Mom or Dad, then brothers and sisters -- and eventually thousands of faces. "It's a monumental task," says Cottrell. "But the brain is equipped to handle it."

How does the brain do it? About one third of the organ is devoted to vision, says neuroscientist Marty Sereno at the University of California at San Diego. One section of the brain's cortex (gray outer layer of tissue) called the fusiform face area specifically recognizes faces. "There are 100 million cells involved in vision," says Cottrell, "and 1 million nerves linking the eye and the brain." Cells near each other in the retina eventually connect to cells in the primary visual cortex. "If someone presented a letter A to your eye, then opened up the back of your head where neurons are firing, they'd see a kind of fuzzy A on the back of your brain," says Cottrell.

As a visual signal flashes up a chain of cells, parts of an image become lines and curves, then shapes, like an eye or nose, and finally an entire face -- all in about 20 milliseconds, or 10 times faster than the blink of an eye. The cells that fire are different for each object or shape you see -- cells in the fusiform face area fire only in response to a face: "We've found cells that are activated only when you see toilet brushes and houses," says Cottrell. But it's the process of detecting subtle differences in the details of a face that scientists hope to perfect in face-ID technology.

DIGITAL DIMPLES

Facial-recognition programs are a series of calculations that tell computers how to "see" faces -- for without proper programming, a computer can't distinguish between a celebrity and a rock. First, however, programmers must enter a database (electronic storage space) of faces into the computer -- the computer needs a stored photo of a particular person to make an ID. "Facial recognition can't find anyone if we don't have an image of them or know who they are," says Prout. College-testing centers could use a database of drivers-license photos, for example, to ensure no one attempts to boost a score using a smarter substitute: "At some point you might have your picture taken for verification when you show up to take the SAT," says Schmitt.

Next, a video camera -- the computer's eye -- must locate a human face in a room cluttered with objects. "This takes pretty sophisticated software, because it's the most difficult part of the process," Schmitt says. The software struggles to accomplish mechanically what the primary visual cortex does naturally: find edges and shapes that define an object.

When a computer pinpoints a face, a different program translates human physical features into a mathematical code, then seeks for a match in the database of coded faces. Each kind of facial-recognition uses a different algorithm (step-by-step procedure for solving a problem) in order to match a face. Software made by Identix in New Jersey, for example, uses local feature analysis, to compare faces. "The human face contains landmarks just like any city," says Identix president Joseph Atick. The software calculates distances between points on a face to create a mathematical map unique to each person.

Eigenface software uses model faces made by morphing common features from photos taken of people across the U.S. Instead of comparing the facial map directly to the database, the software first matches it to one of 128 models to narrow the field and speed up the process.

So far, computer-recognition systems have achieved only 5 to 10 percent of the accuracy of the human eye. "A human can recognize a feature you might not pick up from a computer," says Schmitt: The eye uses clues like scars and freckles that computers aren't programmed to recognize. And the task only gets harder if a person's image isn't exactly the same as his or her picture in the database -- unlike a computer, human vision isn't confused by a titled head, smile or frown, or aging.

Still, facial recognition technology is fast improving. "Computers don't play sports or date, so they can devote all of their 'brain power' to storing faces," says Cottrell. Within five years, he estimates: "The human is going to be left behind."