In search for a suspect, high tech takes over

With reports that a possible suspect in the Boston Marathon bombing may have been pinpointed in photos taken at the scene, facial-recognition technology may be one of the most powerful tools investigators have at their disposal to attach a name to the image.

Counterterrorism specialists said authorities would aim to match the faces of any possible suspects against an array of databases for visas, passports, and driver's licenses in an effort to determine exactly who was in the area when two bombs exploded near the finish line, killing three people and injuring at least 180.

In criminal investigations small and large, one lead can change everything. And a potential match with facial-recognition software can be a better lead than most, providing authorities with the name, physical description, address, and criminal record of a possible suspect.

That information can then serve as a jumping-off point for investigators, but not as conclusive evidence, said Maj. Andrew Ashmar, director of the bureau of criminal intelligence for the Pennsylvania State Police.

"It's not fingerprints," he said. "It's not DNA. But what it does do is provide an investigative lead to follow."

Though facial-recognition programs have been in use for almost 50 years, recent technological advances have led to an increased use by law enforcement agencies around the world. And although it's typically used to help make arrests, the technology can help locate witnesses, identify victims, or, as in a recent Delaware County case, help exonerate someone wrongfully charged with a crime.

In the local case, Tahmir Craig, 23, was arrested in the May 2012 slaying of a minister's son in Chester. Craig, of Chester, maintained his innocence but was jailed while awaiting trial. After a witness recanted a statement implicating Craig, the Delaware County District Attorney's Office asked the FBI to do a forensic analysis of surveillance from the scene. That investigation, which included the use of facial-recognition software, concluded Craig was not the man seen in the surveillance video. Craig was freed last month.

The science of getting computers to "recognize" faces relies on calculations used to map features from a photographic image, including the size, shape, and relative distance between features. Newer systems use three-dimensional sensors that collect geometric information from a photograph, then create a 3-D model that can be repositioned to get the closest possible match.

"Once you have a 3-D model, you can start to manipulate it, change it," said Paul Schuepp, president of Animetrics, a New Hampshire-based company that supplies facial-recognition technology to law enforcement and military intelligence agencies. "The older systems needed a straight-facing photo to get a good match, but we all know that faces look different from a number of angles."

Once a video or photo image is uploaded, that image is then converted to numbers, Schuepp said. From there it's simply a matter of running the numbers through a computer program to see if they come close to matching the numerical map of any other face on record.

"With DNA, you look for patterns," he said. "With the face, you're looking for closeness of features, things of that nature, and by doing that you end up eliminating tons of faces."

In Pennsylvania, the technology is used on a near-daily basis by investigators in Philadelphia and at the state police's Criminal Intelligence Center in Harrisburg, where intelligence and research analysts provide help to criminal justice agencies throughout the state, Ashmar said.

Once a department sends in a photo of a suspect, the analysts run it through the state driver's license system, as well as a criminal arrest photo database. They do not use social networking sites, Ashmar said.

If a submitted photo comes up with potential matches, it returns them in ranking order. Investigators are left with a series of photographs, similar to the photo arrays that police sometimes use when questioning witnesses.

Some matches can be discounted immediately, if there is height discrepancy, for instance.

Near-matches, however, might lead investigators to take a closer look at an individual, particularly if they live near the crime scene, Ashmar said.

Police in some areas have also used it to identify victims of some crimes, as well as witnesses. In theory, it could be used to identify a murder victim found without identification, Ashmar said.

"There are a lot of possibilities," he said, not all crime-related.

Last year, for instance, a photo album was found on the side of a road in Lehigh County and turned over to police.

The album was filled with pictures from a family's Disney vacation. An officer, fearing no one would know where to look for it, scanned a photograph and sent it to the state police.

"Within minutes, we'd ID'd the owner from the license database," Ashmar said. "It turned out they had been in the process of moving, and someone left it on the roof of the car. They were glad to get it back!"