Law Enforcement Facial Recognition
By Michael Ross
This article was published by ComputorEdge, issue #2449, 2006-12-08, as the cover article, in both their print edition (on pages 14 and 16) and their website.
The increasing use of the Internet by criminals, terrorists, and other miscreants certainly has not slowed down their old-fashioned misdeeds in the real world. Bank robbers are still driving getaway cars, pedophiles are still strolling past playgrounds, and fugitives are still hurrying through airport and bus terminals.
One thing all of these bad guys share — even the online crooks plying their craft in cyberspace — is a need to interact with bank tellers, grocers, gas station attendants, accomplices, and all the other occupants of "meatspace". Just like the rest of us, they usually have to visit the bank and the convenience store, even though their method of "payment" is not always legal.
Another thing all of these criminals share, is a face, or at least the semblance thereof. To see just a sampling of the more "popular" ones, check out the FBI's Ten Most Wanted Fugitives. For some more serious-looking mugs, take a look at America's most wanted terrorists.
Many of these faces are unmistakable (to put it gently), and this is precisely what law enforcement officials in the United States and other countries hope to capitalize upon, employing automated facial recognition. Systems using this technology are designed to process real-time snapshots and video feeds of people in public areas, and automatically check each image against a database of mug shots and surveillance photos of known criminals and terrorists.
Not Child's Play
What is one activity that is easily accomplished by an infant, and yet still difficult for even the most advanced programs running on a supercomputer? Recognizing a human face, particularly after minor changes have been made to its appearance, such as the addition of eyeglasses, or a shadow falling on a portion of the face. It becomes even more challenging for a computer, sometimes quite impossible, if the changes are more drastic, such as the addition of a full beard.
Admittedly, this may be an unfair comparison, since the human brain is considered by cognitive scientists to be the consummate pattern recognition engine. One obvious example of this is the phenomenal skill of grandmaster chess players in remembering and recognizing countless patterns on the chess board.
This universal capability is especially true for recognizing people's faces. It is as if humans are hardwired for correctly identifying other humans, as well as their moods and intentions — the latter being essential for survival throughout most of human history.
Regardless of the technical problems faced by developers of facial recognition software, they have made tremendous progress in this area, to the point now where such systems are being deployed for a wide range of security purposes. For instance, facial recognition is now joining other forms of biometric identification — such as real-time fingerprint and voice analysis and matching — for allowing or denying access to restricted facilities.
Recognizing the Dangerous Ones
Automated facial identification is a technology being increasingly considered by law enforcement agencies around the world, for surveillance of much wider areas than ever before possible. For example, police can install cameras in areas of high pedestrian traffic, and attempt to spot known criminals and parole violators, as well as track the movements of suspects in real time.
Facial recognition utilizes various methods for identifying individuals, based on unique physical characteristics. The actual matching process is done using sophisticated algorithms, which convert the raw lines and other shapes into descriptive data, just as an aerial satellite of a forest might be converted into a topographic map, and then into vector data describing the elevation lines of that map.
As one might expect for such a computationally challenging problem, different system developers use a variety of techniques. Some use masks or frames, to which facial features are linked, while others use the distances between particular facial features, such as the distance between the two eyes. All of these complex calculations take time, and thus high performance computer processors are preferred, especially when processing must be done in real time.
To learn more about the state-of-the-art in facial recognition, check the websites for some of the leading vendors, such as Viisage Technology and Identix. They provide some details on the system solutions that they offer to consumer financial service companies, border patrol agencies, and casinos — to name just a few of the types of organizations that are becoming increasingly interested in deploying these systems.
Recognizing the Dangers
Despite the obvious advantages of automated surveillance, it is only valuable if it works. Furthermore, when it fails — such as generating false positives and false negatives — then it can cause more trouble than it is worth.
For example, in August of 2003, the Tampa Police Department abandoned its facial recognition program, because no arrests were made using it; but a fair number of incorrect matches were made. Their program was a testing ground, in that they were the first city in the country to attempt it, and had installed 36 cameras in Ybor City, with the intention of identifying known criminals.
Other factors involved in scrapping the Ybor City program likely included criticism and protests from citizens and the ACLU. This police department was not alone, because at least two airports in the country also decided to discontinue their facial recognition projects.
Proponents of facial recognition for law enforcement purposes, point out that most of the project failures resulted from the difficulties faced by these systems when dealing with variables such as lighting, camera angles, and facial changes. Results greater than 90% accuracy can apparently be achieved in controlled settings with cooperative subjects.
Perhaps a safer approach is to initially limit the uses of biometric identification. Several agencies in Colorado and California (including the San Diego Sheriff's Department) have at one time or another made effective use of full-hand scanning.
Critics of widespread surveillance, including the Electronic Privacy Information Center, warn that this technology may be moving us steadily and rapidly towards a society in which only the most elite citizens will be able to enjoy any privacy. Proponents argue that facial recognition is merely a tool, which can be used for good or ill. Critics may agree with that, but point out that the government's track record in using such tools against their own citizens, is enough to warrant our greatest caution and scrutiny.
Regardless of who wins the theoretical debates and the policy battles, we may see a day when the hidden surveillance camera on every street corner can ask the same question as the curious infant — Who are you? — and discover the answer with blinding speed.
Copyright © 2006 Michael J. Ross. All rights reserved.