The intelligence community is pushing to make biometrically enabled intelligence — the art of identifying people by fingerprints, digital mugshots, iris scans or DNA — a regular part of business.
Biometrics has evolved dramatically during the Iraq and Afghanistan conflicts, and is used in countless manhunts, but the discipline has not been fully institutionalized into the intelligence community. Among other issues, various agencies, including the DIA and the FBI, have different ways of looking at the technological breakthroughs.
“The general field or trade of identity intelligence really is in its infancy,” FBI biometrics expert David Cuthbertson told an audience in February.
Right now, biometrics still mostly means fingerprints. That’s the telltale record most commonly left behind by bomb-makers and thieves. The databases of fingerprints are vast: The FBI has 110 million fingerprint records; the Defense Department, 9.5 million; and the Department of Homeland Security, 156 million.
But other technologies are coming online. Facial recognition algorithms could someday riffle through mugshot databases to find matches much as fingerprint algorithms do today. Iris-matching technology is another field under development. Authorities around the world are rapidly switching from fingerprints to iris scans for verifying the identities of travelers and workers, and iris databases are growing. And some biometrics experts are aiming for multimodal biometrics in which fingerprint matches would be combined with facial recognition and other measurements to determine someone’s identity with maximum confidence.
“Increasingly, in the commercial sector and national, state, and local levels, organizations are going to adopt a biometric capability, like an Iris Scan, to confirm an identity because of its reliability,” a Defense Intelligence Agency spokesman said in an emailed statement. DIA officials turned down interview requests.
But the new choices and opportunities are also creating new problems and conflicting priorities. For example, iris scans may prove more convenient, faster and effective for authentication — identifying travelers, workers and officials — but fingerprints remain the standard for forensics. Some worry that as enthusiasm grows for using irises as ID, fingerprint collection may suffer, leaving fewer troves to search through in an investigation.
In October, DIA opened an Identity Intelligence Project Office.
“As we approach the war’s end, we have laid the foundation for fused capabilities in the Forensic and Biometric arenas and have developed strong interagency partnerships,” the DIA spokesman wrote.
Dalton Jones, a retired Army colonel who oversees forensics, biometrics and identity intelligence at DIA, challenged the biometrics community to look for innovative, cheaper technologies to ensure that biometrics and forensics intelligence “continue to mature and endure as robust, force multiplying capabilities in DoD.” He has warned of the“Threat of Disappearing Capabilities” in a “resource constrained environment,” in a briefing available online.
Over the last decade, biometrics has been put to use for improvised explosive device forensics and for identifying and targeting suspected insurgents and terrorists.
“We’ve always used this information for battlefield-type things, where we feed the special operators this information,” said Don Salo, who directs the Pentagon’s Defense Forensics and Biometrics Agency. “We also need to look at bringing the intelligence community into our fold, which we have now started to do.”
Salo’s agency includes the Biometrics Identity Management Agency, which is the steward of the Pentagon’s fingerprint database called the Automated Biometrics Identification System. He and Cuthbertson appeared together at an Armed Forces Communication Electronics Association meeting about biometrics.
BETTER FACIAL RECOGNITION
One of the fast-growing forms of biometrics is facial recognition, where advances have largely been driven by the private sector. Facebook, Google and Apple are starting to use such algorithms in their social media and photo management software. Advocates say facial recognition will be the way of the future.
“Within, say, five years you probably could develop algorithms that are as robust as your average human to do facial recognition,” said Jonathan Phillips, a facial recognition expert at the National Institute of Standards and Technology.
Vendors flock to NIST to have their facial recognition algorithms evaluated. Phillips said about 20 companies and universities are involved in the latest Facial Recognition Vendor Test. NIST assesses the ability of the software to quickly find matching mug shots in a database.
In the early days of facial recognition software, vendors tended to exaggerate their accuracy claims, so in 1993, NIST developed a standard process for evaluating them. These days, the algorithms are run against a database of mug shots provided by the State Department.
Over the decades, “what we found was that the error halves every two years,” Phillips said.
Still, facial recognition software remains short of its potential. The best accuracy NIST has seen is 92 percent.
One of the largest attempts at a multimodal system, the FBI’s $1.2 billion Next Generation Identification system, is meant to use facial recognition and fingerprints together.
The FBI and its contractor, Lockheed Martin, report great strides on fingerprint matching. In 2011, NGI demonstrated a fingerprint matching accuracy of 99.6 percent, according to Lockheed’s Art Ibers, who runs the company’s NGI work.
“They ran the two systems side by side, and our system found 900 candidates that their system, the legacy system, missed,” Ibers said.
For facial recognition, the performance has improved enough that the FBI decided to run a facial recognition pilot project involving U.S. states. Which ones, the FBI won’t say — “at the states’ request,” spokesman Stephen Fischer said. The subcontractor handling the effort is MorphoTrust USA.
By June 2014, the FBI wants to have in place a better version of its Interstate Photo System, whose mug shot repository is “under-utilized” — in the words of an FBI document — because it’s hard for analysts and law enforcement officers to search for photos.
The repository can only be searched by entering a name or identification number, according to an FBI “Privacy Impact Assessment.” The limitations to that are obvious: Unless one has a name or number associated with a photo already, it’s tough to find a mug shot.
The FBI wants to feed mug shots into the system and search for matches using facial recognition algorithms.
The FBI’s NGI program brought some big changes. The first step started with new desktop computers for analysts and workers at the FBI’s Clarksburg, W.Va., site, where the massive identification database is stored. Next, a new fingerprint-matching algorithm was loaded, tested and approved.
Increment 2 added a remarkable new capability: fingerprinting in the field. Officers with hand-held fingerprint collectors can fingerprint a suspect, and check the identity across the FBI’s Repository of Individuals of Special Concern.
“We’ll return back in a very short amount of time a response that says, hit, possible hit or not a hit,” Ibers said. “That officer will immediately know whether he or she’s dealing with someone who could be dangerous.”
Increment 3, completed in March, introduces automated latent fingerprint and palm processing, so a partial print, in some cases, can be matched to prints in the FBI’s database.
“That is, as you can imagine, a much more challenging problem,” Ibers said.
Lockheed is working on the final increment: the facial recognition part. That is eventually meant to allow more than 18,000 law enforcement offices around the country to launch facial recognition searches across its mug-shot database.
The privacy and collection issues with iris-scan technology mean that, for now, growing reliance on iris scans for identity authentication creates a thorny problem for intelligence analysts and law enforcement officers. When a crime is committed or a safe house is raided, irises aren’t left at the scene the way fingerprints are, and so they can’t be searched against, even if the legal authorities permitted it.
On top of that, there are privacy concerns about storing iris scans of American citizens.
“This trend will result in corresponding concerns over privacy and that one’s biometric identity is privileged information, like a Social Security number, that needs specific protections,” the DIA said.
The FBI will explore iris matching under a separate pilot study set to begin in 2014.
Analysts and investigators prefer facial recognition technology. The proliferation of cameras around the world means that, in a sense, a suspicious person can leave facial evidence behind in all sorts of places.
The problem is that today’s facial recognition algorithms can only find matches from a photo taken straight on and in good lighting conditions.
“There’s a hope that, in the future, improvement can be done to face [recognition] under less constrained conditions,” Phillips said.
The goal is for any image of a face that can be identified with confidence by a human can be identifiable by an algorithm. How does the industry plan to get there? NIST’s Phillips is not privy to exactly how today’s algorithms work. The software is highly proprietary because of its potential commercial value. But Phillips has a general idea about how designers have managed to make so much progress.
They’ve gotten away from attempting to precisely map and compare the geometry of two faces to find a match. They’re making algorithms that work more like the human mind. Subtler clues can give away identity, such as the roughness of the skin, the overall shape of the face, and the general positions of features: “Well, the nose is sort of here, and this is what the nose looks like,” Phillips said.
The job is somewhat tougher for algorithms than for people because algorithms must be cautious. Algorithms are designed to focus on the face and ignore clues that could be changed, such as hair color.
“When you recognize a person in a face image, you use all the information that is available. You use body information. You use hair, eye color. You integrate all the information together,” Phillips said. “The algorithms, for very legitimate reasons, have focused on the interior of the face,” he said.
Changes in poses, such as squinting, or a face at angle, will always be difficult. Human analysts might be called in to work with the algorithms in such cases.
(This story appeared in the May issue of C4ISR Journal.)
Ben Iannotta is the editor of Deep Dive Intelligence (www.deepdiveintel.com/)