Number 245 - October 2003

Facial Recognition
by George Harding, Tucson Computer Society
One of the improved methods of differentiating one individual from another is facial recognition. Others are fingerprints, voice recognition, iris scan, voiceprint, hand scan, retina scan and signature. Each of these can be used in different ways, but the primary use is to determine if an individual is the correct one. If so identified, he/she is admitted or otherwise allowed entry.    Facial recognition has received publicity because it was felt that this method could have identified the September 11 terrorists had the technology been in place. The US Government tested several software packages that are currently available for such identification, to clarify what level of capability they actually have. The results were not promising. First of all, it's necessary to differentiate between two functions of any biometric solution: identification and watch list. Identification is simply that; when a face is presented, the software must	successfully pick that individual out of a database of faces and tell whom he/she is. Watch list is somewhat different; here the task is to pick an individual out of a crowd as being on a list of faces, or not. Watch list mode is what was thought to have possibly caught the September 11 terrorists.    In watch list mode, the software had an average success rate of 77%, using a list of 25 faces. That is, a wanted face in the crowd was correctly picked out 3 out of 4 times. Thus, 25% of the time, the "wanted: individual passed without being apprehended. Further, the false identification rate was about 1%. And that was with a list of only 25 faces; how about 3,000 faces? The success rate dropped to 56%.    It's unlikely that facial recognition will provide much help in identifying terrorists or illegal border crossers, at least with the current state of the art.
Number 245 - October 2003