[Dave Birch] An NIST report
shows that machines can out-perform humans at face recognition under certain circumstances. The FRVT 2006, for the first time, integrated measuring human face recognition capability into an evaluation of face recognition technologies and the performance of humans and computers was compared on the same set of images. The experiment found that algorithms are capable of human performance levels, and that at false accept rates in the range of 0.05, machines can out-perform humans. Note also that one of the reports findings is that the performance iris, face and 3D face is comparable when all three biometrics are acquired under controlled illumination
(my italics). Much to the joy of the Bouncer’s Union, however, it seems that machines are unlikely to replace nightclub doormen any time soon.
Technorati Tags: biometrics
The human visual system contains a very robust and very fast face recognition capability that is excellent at recognizing familiar faces: the nightclub doorman can spot a regular customer, at some distance and in very poor lighting. However, human face recognition capabilities on unfamiliar faces fall far short of this impressive level of perfomance. So if you give the doorman a picture of convicted murderers, say, and ask the doorman to keep them out of the club then the false accept rate (FAR) will be unacceptable.
These opinions are my own (I think) and are presented solely in my capacity as an interested member of the general public [posted with ecto]