Biometrics Expert Delivers Lecture on Facial Recognition
Dr. James Wayman, a nationally recognized expert on biometric technologies, gave a lecture on biometric facial recognition at RAND's Arlington, Virginia office on March 15, 2001. Participants included RAND staff and representatives from the Department of Defense Biometrics Management Office and the United States Secret Service.
Wayman, a member of the faculty of San Jose State University, began by discussing his work for TSWG, the Technical Support Working Group of DoD's Combating Terrorism Technology Support Program. Wayman explained that TSWG is responsible for conducting the national R&D program for combating terrorism through "rapid research, development and prototyping." One of the major missions of the group is the "detection, identification, and surveillance of terrorists" - a task in which biometric facial recognition could play an important part. The publicity that resulted after the use of this technology at Super Bowl XXXV in Tampa, Florida has spurred public debate about the effectiveness of it, as well as broader policy and philosophical concerns about it.
Wayman stated that although human beings can perform facial recognition processes with relatively high fidelity and at long distances, these activities are still very challenging for technological systems. At the most basic level, even detecting whether a face is present in a given electronic photograph is a difficult technical problem. Unless the photograph is captured under very controlled conditions, with each subject looking directly into the camera and filling the area of the photo completely, the system must have ways to decide which round shapes in a photograph are faces and which are not.
If faces are detected in a given captured photo, the recognition system "normalizes" them to match the size and color of the faces stored in its database before matching can occur. The actual matching process is performed by decomposing each face into a weighted sum of component "eigen-faces" and looking not for faces that "look alike," but for those which can be represented by similar sums of these component images.
At their current level of development, facial recognition systems show promise but are not yet advanced enough to be considered mature technologies. Wayman described the recently released "Facial Recognition Vendor Test 2000" study sponsored by the DoD Counterdrug Technology Development Program Office, DARPA, and the National Institute of Justice. This comprehensive study of current facial recognition technologies showed that differences in factors like camera angle, direction of lighting, facial expression and other "minor" parameters can have significant effects on the ability of systems to recognize individuals. Basic probability calculations showed that even in systems with a very small "mug shot database," in order to produce even a system where the probability of misidentifying individuals was 1 in 20 (i.e., there was a 1 in 20 chance of falsely matching a sampled face to a "mug shot" in the system's database), there would be about a one in two chance of a person not being matched to their own picture.
Wayman concluded by saying that recent tests of these technologies clearly indicate that the current capabilities of facial recognition systems are limited. There is a great deal of room for improvement in both the algorithms used to match sampled faces and in databases of file images. The moderate level of success that current systems have displayed must be placed in the appropriate context, however; while human beings can often readily recognize faces at long distances, the efficiency of such recognition falls precipitously when posted human guards are asked to scrutinize large crowds in search of small numbers of potentially threatening individuals. As a result, for these tasks, the current technical capabilities may even exceed more traditional approaches, and combinations of automated and human-based recognition could be advantageous.
It is of critical importance that the capabilities of systems and potential ways of applying those capabilities are appropriately matched to security and surveillance needs so that individuals neither expect too much or too little from these emerging technological tools.
Following Wayman's technical presentation, John D. Woodward, Jr., a member of RAND's Biometrics Team, gave an overview of the relevant law and policy framework surrounding the use of biometric facial recognition in public settings. Woodward explained that the United States Supreme Court has determined that government action constitutes a search where it invades a person's reasonable expectation of privacy. But the Court has found that a person does not have a reasonable expectation of privacy in those physical characteristics that are constantly exposed to the public, such as one's facial characteristics, voice, and handwriting.
Therefore, although the Fourth Amendment requires that a search conducted by government actors be "reasonable," which generally means that individualized suspicion is required, the use of these technologies does not constitute a search. As a result, while the use of facial recognition at the Super Bowl has generated a significant level of public discussion on this topic, such an application is allowed under the current legal framework surrounding these activities.
To learn more about law and policy concerns of biometric facial recognition, see "And Now, the Good Side of Facial Profiling".