How facial recognition helped police identify the Capital Gazette shooterJune 29, 2018
In the hours after yesterday’s deadly Capital Gazette shooting, one critical fact remained elusive: the shooter’s identity. Reports suggest that Jarrod Ramos did everything in his power to avoid being named and that it was only through the use of facial recognition technology that police finally identified him.
Ramos had no identification on him at the time of the attack, and soon after his arrest rumors began to surface that he had filed off his fingerprints to prevent identification. Police have backed away from the claim since, saying only that the initial attempt to fingerprint him had failed. In a press conference today, Anne Arundel County Police Chief Timothy Altomare said that “we had lag getting answers on fingerprints is all I can tell you” and suggested that the technology might be to blame. “I don’t know why it took us a little longer on the computer system. Computer systems do that,” he said.
That’s not altogether surprising: fingerprinting failures are relatively common, with some industry estimates putting the error rate as high as ten percent. Some of that error rate comes from poor equipment or fingerprinting technique, but a significant portion also comes from problems with the prints themselves. A person’s fingerprints are only as durable as their skin, and they can be rendered unreadable by anything from manual labor to extreme swelling. Experts typically hedge against this by printing multiple fingers at once and relying on optical techniques that are more capable of dealing with abrasion. There’s ongoing debate over how difficult it really is to read a print from damaged fingers, but it’s a very real problem and there’s nothing necessarily ominous about a failure to read a print.
With no fingerprints and Ramos refusing to cooperate, police turned to more creative methods. According to multiple reports, police used facial recognition technology to identify Ramos. Altomare told reporters that police sent an image to the Maryland Combined Analysis Center, and that the center compared it against images in the Maryland Image Reposity System (or MIRS), which includes over ten million photos drawn from known offenders and the state’s entire driver’s license database. The system also has access to the FBI’s mug shot database, which would add another 25 million faces to search through. Because Ramos was formerly charged and convicted for harassment, it’s likely that his mug shot would have appeared in at least one of these databases.
This isn’t the first time that the Maryland police department used MIRS photos to identify a criminal. Earlier this month, Maryland police caught a robbery suspect by comparing an Instagram photo against the state DMV database, which ultimately led them to his name and driver’s license. Many states, including Maryland, already use recognition software at the DMV as a way to spot duplicates and potentially fraudulent identities in the system. Because both photos are taken under the same controlled conditions, that kind of recognition is significantly easier than performing facial recognition from surveillance footage.
While the method seems to have performed well in Ramos’s case, there are still significant civil liberties concerns around the way police use MIRS. In the press conference, Altomare admitted that the system had “come under some fire” from privacy advocates. Police are supposed to remove people who were arrested but found innocent, but since the system is rarely audited, it’s hard to say if that’s actually happening. There are also racial justice concerns, given racial disparities in rates of arrests, compounded by higher error rates for African Americans in many facial recognition algorithms.
Still, the system used by the Maryland police seems to have been effective in this case, correctly identifying a culprit when traditional methods failed. Altomare added that identifying the attacked would have taken much longer without the facial recognition system in place. “It was a huge win for us.”
Shoshana Wodinsky and Adi Robertson contributed reporting to this article.