FBI Facial Recognition System Gives Officers an Investigative Lead

The powerful tool replaces legacy technology and lets police officers automatically compare a suspect's digital facial image against more than 20 million images, but it has accuracy limits and has raised concerns among privacy groups.

by / October 20, 2014
FBI CJIS Division

New FBI facial recognition technology released in September means more than 18,000 law enforcement agencies can search potential criminals by face in addition to fingerprint. 

The facial recognition tool, called the Interstate Photo System, lets officers automatically compare a suspect's digital facial image against the 20 million and growing images available for searches, giving officers an investigative lead.  

"What this does for our criminal justice community is it provides them another tool to be able to go out and identify criminals," said Stephen Morris, assistant director of the Criminal Justice Information Services Division of the FBI. 

The facial recognition tool is part of CJIS' Next Generation Identification (NGI) system, which is a 10-year IT project begun in 2008 to replace the decades-old legacy Integrated Automated Fingerprint Identification System. The project has been launched in stages, but the September release marks the biggest rollout and the official end of the legacy system. The facial recognition technology was piloted in six states and developed in collaboration with law enforcement agencies nationwide. NGI currently operates in about 75 percent of the country's law enforcement agencies.  

"This is a long overdue effort to replace legacy technology, old technology, with new, relevant, more efficient, cheaper technology and, more importantly, more accurate technology," Morris said. 

The facial recognition technology represents the first time officers can search CJIS' criminal mug shot database, which can store up to 92 million photos, against digital photos culled from investigations. Previously, there was no way to automatically search against the images collected along with fingerprints taken during booking or incarceration. Law enforcement officers would have to submit photos to the CJIS Division for facial recognition processing. With the new system, officers can choose between two and 50 candidates for review.    

Any image used for search purposes is in law enforcement's possession pursuant to a lawful investigation, Morris said. Digital photos, for example, can be taken from surveillance cameras or from digital devices that are seized with a search warrant. The ability to use the images captured on these devices is where the value in the tool lies, he said.  

"Obviously you can't pull a fingerprint off of a phone, but if there are images on a phone and you know that it's that person's phone, it's the next best thing," said Morris. 

Facial recognition technology, however, is less reliable than fingerprint identification, with the Interstate Photo System returning the correct candidate a minimum of 85 percent of the time when a matching photos is in the repository. Any facial recognition hits are therefore investigative leads, not positive identifications, Morris said. 

"In other words, it's not an absolute identification," Morris said. "When that agency gets that result back, they then have to go out and do the follow-up investigation."

Additionally, controlled environments are best for facial recognition, a relatively young technology, which can be explained as an algorithm that make sense of millions of pixels describing facial features, said Chenjgun Liu, associate professor of computer science at the New Jersey Institute of Technology. DMV photos, for instance, are a good use for the technology. 

"There is no such thing as a system or program that can recognize people without any constraint," Liu said. "That is a fiction."  

Added NGI Capabilities

The Next Generation Identification System overhaul of CJIS' former fingerprint legacy system has included many new capabilities for the FIBI, including the rollout of the Advanced Fingerprint Identification Technology in February 2011, increasing CJIS' fingerprint-matching accuracy to a rate of 99.6-percent from 92 percent. 

Other NGI System services that are also available include a mobile fingerprint search, enhanced latent print capabilities and repositories for searching palm prints, and enhanced processing speed for the entire system.  

Another capability released in the latest rollout is Rap Back, which was developed for organizations requiring civil fingerprints for background checks and licensing that have elected to be notified regarding any subsequent activity on a person's record, in addition to that first snapshot. 

Liu, who has received funding from the Department of Defense to support his research into improving the technology, recognizes the benefit of using the technology with digital images to narrow down the number of suspects in an investigation, reducing the search effort dramatically. 

"The potential benefit is of course also obvious. We have nowadays images almost everywhere," he said. 

CJIS has put into place specifications to ensure photo quality for people submitting digital images to its database, requiring they be frontal facial images with no shadows, and be taken in controlled environments. The accuracy of the photos both in the database and for those that are searched against are correlated with faster and more accurate search results, Morris said.

And although it's not an absolute, searching for both photo and fingerprint matches for one person can give officers almost virtual certainty of someone's identify, he said.  

"For the folks out there worried about it falsely identifying people, I would say it actually closes the gap and reduces the chance of an individual being falsely identified," Morris said. 

Indeed, the fact that law enforcement can search against such a large database of digital images, has some groups uncomfortable with its possible surveillance capabilities. The facial recognition technology received attention in the spring from a Freedom of Information Act lawsuit filed by the Electronic Frontier Foundation.

Morris said NGI has been subject to privacy threat assessments and privacy impact assessments, and that abuse of the technology using photos on social networking sites is "patently false." 

"First and foremost all of these things are done with absolute guarantee that privacy and civil liberties are of first concern," he said.  

Cost of the technology overhaul is another concern. The entire NGI System is a billion-dollar project. But Morris said the high price tag is an investment. "Over a long run, over a 20-year span, the return on that will be significant. You're talking about savings in the hundreds of millions of dollars." 

That's because the technology was built on a flexible framework, scalable as new biometric capabilities become economically and technically feasible. One such technology, iris image recognition, was just piloted under NGI. 

Although the technology is not ready to be added to NGI's set of biometrics tools, it soon may be, Morris said, just as facial recognition technology has come around. 

Jessica Hughes Contributing Writer

Jessica Hughes is a regular contributor to Government Technology and Emergency Management magazines.