Computer Puts Names to Faces of Criminals

It’s a scene straight out of film noir, yet it’s played out in real life many times a day: A beleaguered crime victim flips through page after page of mug shots, looking for his attacker, his head swimming amid the hundreds of faces glaring back at him.

But using a program that’s the first of its kind, Santa Ana police hope to have a computer do that search instead, saving crime victims--and themselves--countless hours.

The Facematcher program, developed by a Fountain Valley software company, allows the police to take an artist’s sketch of a suspect or a still from a surveillance video and compare it against more than 40,000 mug shots in their database.

As never before, the computer can match a name to the face, Capt. Dan McCoy said.

“Instead of having a person like in the old movies going through mug book after mug book, the computer goes through it for you, at 1,000 photos a minute,” said McCoy, who added that the goal is to expand the program, if it succeeds, until it is statewide, and eventually nationwide.

After a few seconds of image- and number-crunching, the computer comes back with a selection of mug shots that it believes most resemble the face of the suspect it was given.

“It doesn’t give you 100%, ‘Here’s the photograph, excluding all others.’ It gives you the highest probabilities,” McCoy said. “It doesn’t mean it’s always going to ID the suspect, but it certainly gives you more of a head start.”

Even if the suspect isn’t among those in the department’s database, the victim can use the photos provided to refine a description--”the eyes were more like this guy’s, and the mouth was more like that one,” said Tom Taverney, vice president of Pacer Infotec, the company that developed the software.

The program will be tested through April by Santa Ana investigators.

The project is being funded by a $250,000 grant from the National Institute of Justice, the research and development agency of the U.S. Department of Justice. Santa Ana was one of only 17 grant recipients among 117 law enforcement agencies that applied for the federal funds, and the only agency using the Facematcher program.

“The fact that it hasn’t been done in any other police department is key,” said Trent DePersia, director of the National Institute of Justice’s research and technology division. The face-matching program “intrigued both the technicians and the practitioners on the review team.”

The program starts to do its work when police scan in a surveillance camera photo or artist’s sketch of a suspect. The computer lays a grid over that face to measure the relationship between its features--the distance and angles between the eyes and nose, for example.

The computer compares the new face to Santa Ana’s mug shots and finds the most likely matches. The screen then shows those 10 or 12 or 100 photos on the screen--however many are requested--in descending order of how closely they resemble the suspect’s image.

The search can be speeded up by limiting the number of mug shots to be compared against the suspect’s image. If the victim knows the suspect was a white male over 6 feet tall, for example, the computer won’t have to search its memory banks of people who don’t fit those criteria.

What is unique about this process, programmers say, is that it mimics the way humans remember faces, by remembering the relationships between the features, rather than committing to memory every square centimeter of a person’s visage.

That’s the reason seemingly crude composite sketches work, Taverney said. The picture doesn’t have to be an exact reproduction as long as its relationships are accurate.

“This is something police have been saying they’ve wanted for years,” he said, but even when it became technically possible it still wasn’t cost-effective until now, with the simplified approach Pacer Infotec has taken.

One method discarded along the way copied the navigation techniques used by cruise missiles, taking three-dimensional snapshots of the surface. But Taverney said such a detailed system wasn’t feasible for face matching--every search required the full attention of a supercomputer costing millions.

In comparison, Santa Ana is going to be running its Facematcher program on a 233-MHz Pentium-powered desktop computer, which anyone could buy at the mall for $2,500.

While working with researchers from a German university, Ruhr-Universitat Bochum, the programmers first had to teach the computer what a face looks like: The nose always goes between the eyes and mouth, the eyes are usually a certain distance between the ears, et cetera.

“If you show it an image of a Cyclops, it isn’t really going to know what to do with that,” said Pacer Infotec’s chief engineer, Eric Olsen.

Eric Sterling, chairman of the Criminal Justice Policy Foundation in Washington, D.C., said, “As a way to solve crimes, [it’s] a very important potential breakthrough.”

But he cited concerns about the system’s reliability and the danger of a victim or witness--wowed or cowed by technology--mistakenly identifying one of the suspects offered by the computer.

“The computer, because it is a computer, isn’t necessarily any more magical or foolproof,” said Sterling, former counsel to the U.S. House Subcommittee on Crime and Criminal Justice. He noted the additional danger that police might then waste time by focusing their investigation on the wrong person.

“You want the data to be reliable, and you want the police response to match the degree of reliability,” he said. “If it’s a basis for further investigation, then that’s a very useful kind of tool.”

McCoy said the system can quickly help police identify someone they’re looking for and will be even more powerful when mug-shot databases are linked, and photos throughout California or even the United States can be searched.

Taverney said such a widespread system could have hastened the capture of the Unabomber suspect, for instance, using the sketch that authorities spent years vainly displaying in the media.

And DePersia said the face-matching technology could be even more effective when used with other means of identification--such as identifying a bank robber from both a photo and from

fingerprints left on the counter.

But Sterling said the public should guard against privacy violations if the system gains widespread use, raising the question of whether high school yearbook photos or driver’s license pictures would be added to the database.

Taverney said many people thought police already had this face-matching capability, but in fact it only existed in science fiction.

In the movie “Robocop,” for example, the title character takes video of an attacker that he’s recorded in his computerized brain, and runs it through the police department’s system. Voila, he matches the video with a mug shot, and finds the corresponding name and other information about the bad guy.

“It changes the way people investigate,” Taverney said of his company’s real-life system. “It speeds things up. It brings people to justice faster. I think it’s pretty groundbreaking, on the order of fingerprint matching or DNA.”

(BEGIN TEXT OF INFOBOX / INFOGRAPHIC)

Criminal Pursuit

The Facemaker computer program compares facial characteristics and compiles a gallery of likely suspects’ mug shots for crime victims to review. Police scan in artist’s sketch or surveillance photograph. The computer lays a grid over the face and measures the facial features. Distance and angles of eyes, nose, mouth and ears are compared by computer and potential matches are made. An example:

Source: Pacer Infotec Inc.; Researched by STEVE CARNEY / Los Angeles Times