FACING THE FUTURE OF RECOGNITION SOFTWARE

Two of the most frequently mentioned names in law enforcement and security circles after the September 11, 2001, terrorist attacks have been Joseph Atick and Thomas Colatosti, Chief Executives, respectively, of Visionics Corporation of Jersey City, New Jersey, and Viisage Technology of Littleton, Massachusetts. The incidents initiated a host of questions about a product they (and their competitors) have long been touting, but with somewhat limited success: Facial Recognition Software (FRS).

FRS is viewed by proponents as a potentially valuable tool in combatting terrorism and reducing routine workloads for law enforcement agencies (LEAs). Others argue that it is not. Both sides have merits to their arguments. In either case, one thing to remember is this: FRS is a relatively new technology and its applications and value have not reached a stage where they have been tested fully.

Developers are still tweaking and enhancing FRS technology. Consequently, its true worth may not be evident immediately; however, it does have potential as a viable tool in security and law enforcement activities, but not until the ongoing “field testing” is completed. Its real value may come when it is integrated with complementary systems (e.g., closed circuit television systems) and the mystery surrounding its disappears.

Exactly how, and where, FRS will be used is a bit mysterious right now, as is the process on which it is based in part: biometrics, i.e., the statistical study of biological data. Currently, FRS is being used in one form or another, on a stand-alone basis, in airports, casinos, government offices, correctional institutions, city streets, police departments, et al. Where it is used is not always revealed.

Not everyone is anxious to disclose the locations where FRS systems are installed. For example, Viisage Technology announced on October 5, 2001, that it had signed a contract with a major U.S. airport to install one of its systems. Company officials did not say which one, though, partially because secrecy can be used to the airport officials’ advantage. However, neither vendors (nor users) are reluctant about proclaiming the benefits of such systems. They are more than willing to proclaim the benefits of FRS and predict a great future for it. Laudatory statements aside, the technology is not without its drawbacks. There are as many cons to it as there are pros. Let’s take a look at both; how the technology works; who uses it; and for what purposes, etc.

The facial recognition process is based on computer scanned pictures. The pictures can come from anywhere a camera might be mounted, e.g., ATMs, casinos, parking lots, traffic intersections, stores, or sidewalks. After all, there are approximately 2,000,000 surveillance type cameras located in the U.S. today, with more being installed all the time. So, the availability of photos is not a problem, although the availability of clear photos suitable for scanning might be.

The effectiveness of an FRS system is only as good as the photos with which it works. The software involved searches through a database of pictures for matches with individuals scanned. Computers process a picture of a face and make scores of measurements, e.g., the distance between the eyes or the angle of the chin. Visionic’s product selects between 14 and 20 facial landmarks, triangulates among them, and calculates the face’s geometry. The process creates about 84 bytes of data which means a modern computer can process it and several other faces simultaneously. Significantly, fooling the system is not easy.

Some detractors suggest that terrorists, criminals et al need only put on glasses, wear a beard or mustache, apply make-up, etc., to alter their appearances. That may deceive the human eye, Atick said, but it won’t fool Visionic’s system, simply because it examines the underlying structure of the face, rather than the attempted disguises. Drastic plastic surgery might work though, Atick conceded. But, such arguments are moot if there are no pictures available in the first place.

The picture of the subject being scanned must be included in the database used, which can be local, regional, national, or even international, depending on who is using the system and for what purposes. That creates another drawback: the lack of suitable photos. As some wags have commented, terrorists are not likely to pose for photos, without which FRS has limited value. Even when pictures of known criminals, terrorists, and other bad guys do exist, they may be too dim, grainy, or otherwise limited to be of any help. Some usable pictures do exist, though.

Atick acknowledged that Visionics had photographs of at least three of the September 11th hijackers, and he added that there was a chance they could have been stopped early if officials had been able to match faces in the crowds with suspects’ mug shots. Colatosti agreed that the technology could have made a difference had it been deployed that day. In either case, the lack of clear, positive photos can - and does - inhibit the efficacy of FRS systems. So, for now, the databases, for the most part, comprise known criminals or, at least, folks who have had contact with law enforcement and security agencies. As a result, “matches” tend to identify only people who are already known to these agencies. That is a good thing as far as civil libertarians are concerned.

Some opponents of FRS believe that its indiscriminate use invades people’s privacy. That may indeed be the case in societies which place a premium on individual freedom. But, it also raises the question of which takes precedence, individual rights or the better good of society. For example, if the crime rate in a city that uses facial recognition technology drops as much as 40 percent in one year after the process is implemented, does that diminish the significance of individuals’ rights, since society as a whole has benefited? That is not a hypothetical situation.

FaceIt®, a system sold by Visionics Corporation, led to a 40 percent drop in crime in the Newham Borough of London, England, within one year after it was installed in the fall of 1998. That led officials to implement a youth crime initiative in Newham based on facial recognition. Moreover, the success pointed to a potential expansion of the system to other areas of the United Kingdom which has a long history of using closed-circuit television systems which can now be integrated with FRS as a crime fighting tool. As more and more facilities, LEAs agencies, etc., implement facial recognition systems, claims of privacy violations will no doubt increase, which is something providers have taken into account.

Atick addressed that issue recently. He noted that his company has always taken an active role in articulating the most appropriate and ethical ways for society to gain benefits from its technology and to ensure that the technology cannot be misused. Thus far, he said, “We have formulated responsible use guidelines; secured their acceptance by those adopting the technology; been vigilant in ensuring compliance; and, where possible, we have built technical measures to maintain control over the installations.” Visionics has gone a step farther.

Atick reported that the company will work closely with federal legislators and privacy interest and industry groups to “share its knowledge, experience, and privacy protection principles pertaining to surveillance and other applications of face recognition technology.” In addition, it will continue to promote an improved public understanding of the creation, use, and control of facial recognition systems and will support public policy which ensures that deployments of facial recognition are conducted in a way that uphold all privacy rights.

Atick offered a few principles to provide the basis for the proposed legislation. Among them are:

• establishment of public communication systems, such as signage and media alerts;

• guidelines for database protocols, e.g., need justification for inclusion and removal, valid duration of information, dissemination, review, disclosure, and sharing;

• nonmatch, no memory, i.e., ensuring that no images are kept by the system unless matched as a criminal;

• technical measures for database size and integrity, as well as authorized operation and access; and

• recourse and penalty for violations of the principles.

Such principles are commendable, but, if implemented, they could lead to excessive fluctuations in databases, which are, to begin with, not 100 percent reliable. They almost need a “perfect world” in which to function effectively.

FRS does not always work well in nonstandard environments, e.g., airports. As noted earlier, the pictures entered into the databases should be as clear as possible if the software is to pick out “matches.” That is not always the case if the pictures were taken in dim or indirect lighting, for instance. Certainly, if the pictures used were taken for driver’s licenses or mug shots (to identify people committing fraud, e.g., welfare recipients) or regulating access to weapons and biohazards, they might be clear enough for FRS databases. Such photos are taken generally under constant lighting. That facilitates the creation of a local database, e.g., for a police department or welfare agency. Indeed, that is how many LEAs create their databases to accommodate their growing numbers of systems.

LEAs are acquiring FRS systems for a variety of purposes. Imagis Technologies, a Canadian developer of image identification software with a focus on biometric facial recognition, has been successful recently in selling its various products as individual modules or integrated systems to a variety of customers. For example, several California LEAs - the Chino, Fullerton, and Porterville Police Departments, and the Guilford, Connecticut, Police Department - have all purchased modules or integrated systems recently. Their acquisitions include Imagis’ CABS Arrest & Booking software and the ID-2000 biometric FRS. The system can be included in a fully integrated computer-aided dispatch system, records, and mobile use for police cruisers. The question, though, is whether the systems can be linked with one another.

The possibility exists. There are four police departments in the West Covina (CA) Service Group linked to one another, and a group spokesperson said that several more agencies in the area will purchase FRS systems in the near future. At this point, there is not much networking going on with FRS systems, primarily because there are not a lot of them up and running. Thus, agencies that acquire them tend to be more focused on building their databases and getting the systems up and running than they are on networking. That is a drawback, at least for the moment.

Another problem with FRS is that early results with installed systems have not been conclusive one way or the other. A number of agencies have tested the technology with disappointing results. For instance, the Department of Defense’s counterdrug technology office ran a test in 2000 which highlighted some of FRS’ weaknesses. It found that (even under carefully controlled conditions) the systems it tested, generally considered to be the best commercial ones available, had only a ninety percent positive rate. Simply translated, that means the systems failed to detect people they were supposed to identify ten percent of the time - and, they incorrectly identified innocent people as wanted suspects approximately one third of the time.

Similarly, the Immigration and Naturalization Service (INS) tested an FRS system to identify people in cars at the Mexican border. That did not work out, either. Consequently, the INS stopped using it. A department spokesperson acknowledged that the system has potential, but not for INS’ immediate use. Other tests have been inconclusive as well.

There was a test conducted at the Super Bowl in Tampa, Florida, in 2001. Viisage technology was used to monitor fans going through the turnstiles. There were nineteen “hits” registered by the system. Some of the people identified were wanted for minor crimes, such as ticket scalping and prostitution; however, not everyone recognized by the system fell into the “most wanted criminal” class. When test participants compared the faces of the people “hit” at the turnstiles and the photos in the database, they did not always match. The number of “false positives” was never released, but the test results demonstrated that FRS is still not at a point where it can be relied on totally. The question, then, is just how reliable is FRS in its current technological stage? And, if a test at a football stadium generates a number of “false positives,” what will happen when a system is implemented at a busy airport through which millions of passengers come and go in a relatively short period of time? That is a question which must be answered if FRS is to become a commonly used security product.

Currently, FRS techniques have a failure rate of approximately 10% to 20%. That is acceptable to some proponents who suggest that the 80% to 90% accuracy rate means the product is ready for implementation. At least it can identify the “enemy,” as Atick suggested. “The cornerstone of our defenses against terrorism is the ability to identify the enemy,” he told reporters. “In previous wars, the threat was presented by a known enemy. They had insignia; they occupied a certain part of the battlefield. Now individuals do not announce themselves as the enemy, and we have to identify them.” FRS can do that - but not all the time. But, the percentages might change soon.

There is a new system being developed by Neurodynamics, a British company which is well-known for its work with pattern recognition software. Neurodynamics is testing a three-dimensional face recognition system which has accuracy rates of 96% to 97%. The system shines an infrared light into the subject’s face and records distortions in the light to create a three-dimensional view of the face. There is a catch to this system, though: It needs a reference, which standard photographs cannot provide.

People who will be included in the database must supply infrared images of their faces against which new photographs can be matched. Once again, criminals, terrorists, and assorted desperadoes may not be willing to supply such photographs which severely reduces the value of the system. Nevertheless, Neurodynamics’ innovation, which has been under development for six years, suggests that improvements in facial recognition systems are on the horizon which is a positive development for law enforcement and security personnel.

Other new developments include the changeover from analog to digital cameras and clearer storage of photos on hard drives. Digital cameras provide sharper images - and can magnify images enough to zoom in on faces of people a mile away. Those sharper pictures can be stored more clearly on hard drives through the use of the technology used on consumer digital video recorders. Those are but two technological enhancements in the works.

All in all, facial recognition software may not be the one answer to security problems exacerbated by the recent terrorist attacks - at least not for now. However, when integrated with other products such as eye, fingerprint, and hand scans, they can be a viable tool. Perhaps, when they are enhanced, they might be effective stand-alone tools. Either way, FRS is helpful as one tool. That, on the face of things, is a positive sign for law enforcement and security personnel.