Biometrics fact or science fiction?
James Wayman was 30 years old when he accepted a Department of Mathematics faculty position at the U.S. Naval Postgraduate School in 1981. As a faculty member, he mentored students, built algorithms, tested speaker-recognition techniques and theorized about a futuristic field that eventually emerged as biometrics.
Wayman now is the director of biometric identification research at the San Jose, Calif.-based U.S. National Biometric Test Center, which advises the U.S. and U.K. governments on the use of biometric identification devices. He also heads an international delegation focused on developing a single vocabulary for researchers who build large-scale, interoperable, government-sponsored systems. Such systems unite speaker recognition, fingerprinting recognition, hand-geometry recognition and facial recognition.
“These fields are united under the definition that biometrics is the automated recognition of individuals based on their biological and behavioral characteristics,” he said.
Fingerprinting systems that emerged from the forensic community are the most acceptable biometric technologies in terms of court admissibility and standardization. Indeed, fingerprinting is the most commonly used biometrics system by public-safety agencies.
In Harris County, Texas, population 3.7 million, sheriff’s deputies use a wireless, hand-held device to capture fingerprint data, said Pete Schroedter, manager of the Harris County Sheriff’s Department’s automated fingerprint identification system (AFIS). He said 47 hand-helds are deployed currently, and he expects to distribute another 30 to officers who serve warrants, hunt fugitives and gather criminal intelligence data.
A Department of Homeland Security grant worth approximately $1.2 million paid for the system, which includes Motorola’s Printrak Biometric Identification back-end solution, its biometric software package and Datastrip’s hand-held device. The device wirelessly transmits encrypted fingerprints an officer obtains while in the field to a centralized criminal database. The database then uses algorithms to search for a match via a completely computer-automated process. If there is a match, the device displays the suspect’s photo and his or her criminal record. A commercial cellular link supports the data transfer.
Schroedter said the automated system increases the efficiency of fingerprint processing and provides real-time data to officers in the field. However, it’s limited to only one search at a time, which created a jammed pipeline effect on his network. Because the department used its primary network to handle the AFIS system’s approximately 700 daily transactions, responses to mobile devices took anywhere from 3 to 5 minutes.
“I can manage the system by priorities,” Schroedter said. “But at what point do I make the mobile systems a higher priority? And if I do that, than am I going to have a backup of requests at the inmate processing center or from people who are trying to file charges?”
To solve the problem, a separate back-end system now supports mobile devices. The system holds roughly 1.5 million records and touts a 1-minute response time from the point where an officer records a fingerprint to when he or she receives identification.
Fingerprinting is the most mature biometric technology, and more devices are integrating it into its systems, said Nick Orlans, principle engineer of biometrics and identity management for the Information Security Center at MITRE — a not-for-profit organization that manages federally funded research for U.S. government agencies. Orlans’ current work focuses on developing computations that decrease the cost and increase the speed and throughput of large fingerprint systems, such as a billion-person database.
However, it has been tough to find a reasonable price point.
“It’s a dollar per match,” he said.
In addition, fingerprinting devices tested in the lab often do not perform as well in the field, failing the reliability test, Orlans said. In an unconstrained environment, Mother Nature controls how digital prints are captured. For example, when devices are used on a cold morning, a warm finger generates halos around the image and makes it unusable.
Still, public-safety agencies are investing in not only fingerprinting devices but also those devices that can capture a suspect’s image. Dan Gomez, operations officer in charge of the Los Angeles Police Department’s (LAPD) tactical technology unit, said his department is in the beginning stages of procuring wireless, facial-recognition devices. He hopes the process is complete by month’s end so he can deploy at least 300 devices citywide before 2008.
Facial-recognition technology in reality is a far cry from its portrayal in Hollywood movies like “Enemy of the State,” where the characters’ images are captured on a convenience-store surveillance camera and an automated computer system displays their identities — and rap sheets — to National Security Agency agents’ hand-held devices.
“That’s science fiction,” Wayman said.
Instead, the LAPD’s system acts more like an electronic mug shot book. Officers with probable cause would use the device to scan a suspect’s face from within a few feet. The video data is compressed and then transmitted to a centralized database. Possible face matches appear on the hand-held, where the digital photograph and the on-file mug shot display side by side. The officer must decide whether it is truly a match.
“If [officers] confirm the match, the data is stored, the perpetrator processed and an arrest is made,” Gomez said. “If the officer reports it is not a match, all the data is deleted from the system. There is just an electronic copy that states a transaction occurred, but no data was stored related to the picture.”
Researchers are determined to bring facial-recognition to the next level. At the NSA, Wayman developed a technology that taught computers to locate people in scenes. Tracking faces and objects — and deciphering between the two — is the first step to next-generation, automated facial recognition, he said.
Henry Schneiderman, CEO of Pittsburgh Pattern Recognition, began researching ways to improve facial-recognition technology in the mid-1990s. He was a student and then faculty member at Carnegie Mellon University’s Robotics Institute under the advisement of Takeo Kanade. According to Kanade, the officer does all the work in facial detection — as evidenced by the LAPD deployment — but that will change as the technology evolves.
“Imagine you have a camera looking at the parking lot, and it must do what the officer did — take the right picture, of the right person, of the right thing — which is the face — at the right time,” Kanade said. “That’s what I call face tracking.”
Pittsburgh Pattern Recognition now licenses the technology from Carnegie Mellon and primarily sells it to government agencies. Its software development kit finds faces in photographs and, in video surveillance footage, tracks subjects’ movements throughout each frame.
An undisclosed U.S. government agency uses the software kit to record all the daily news broadcasts worldwide, Schneiderman said. Agents don’t have the time to sort through video files to determine whether there is an event of significance — a military uprising or an earthquake. Therefore, the software filters the video and isolates human faces and objects.
“It doesn’t yet identify those faces,” Schneiderman said. “It just says, ‘That’s a human face; it’s moving in that direction.’”
One looming issue is whether data gathered on fingerprint or facial-recognition systems is reliable enough to meet the Daubert criteria. The Daubert criteria stems from the 1993 U.S. Supreme Court case Daubert v. Merrell Dow Pharmaceuticals Inc., which sought to clarify how courts used expert testimony and set evidentiary science standards. The decision instructed judges to examine the underlying scientific method they believe to be relevant and reliable.
“Fingerprinting is admissible as scientific, but … that’s our oldest biometric method, and we are just now getting to the point where [the scientific community] agrees that it is indeed scientific,” Wayman said. “[T]hese other technologies are not even close. We are no way able to meet the Daubert criteria.”
Harris County’s Schroedter sees facial recognition as a tough sell in the courtroom. As a latent print examiner for 27 years, he took fingerprints found at a crime scene and compared them to a plaintiff to prove a biometric match.
“The defendant in court has the right to cross-examine the expert witness,” he said. “If the computer is making the match, how are you going to cross-examine that computer? Then, if that computer messes up one time, everything is for naught. All these other people have cases for appeal.”
Wayman said these are all examples of how far the technology needs to go before it meets scientific reliability and legal standards. More important, the scientific community must realize these systems are more than just algorithms and computations.
“We don’t understand enough about controlling people’s behavior in order to really make these systems work,” he said. “In other words, we try to write all this computer code and forget that the human is more than fifty percent of the equation.”