Face recognition gets better

Computers have become much more adept at ‘recognising’ human faces during the past two years according to a new report from the National Institute of Standards and Technology (NIST).

Scientists from NIST, the Defense Advanced Research Projects Agency (DARPA) and the DoD Counterdrug Technology Development Program Office have completed a comprehensive study of commercially available face recognition systems and concluded that the technology has made significant advances since 2000.

The primary objective of the study, dubbed Face Recognition Vendor Test 2002 (FRVT 2002), was to provide performance measures for assessing the capability of automatic face recognition systems to meet real-world scenarios – verification of identity, identification of an unknown individual and detection of an individual on a watch list.

Key findings of the study include new data about verification, demographics, and indoor and outdoor matching abilities.

The study shows that there has been a 50% reduction in error rates since comparable tests were conducted in 2000. For verification (i.e., determining whether a person is who he or she claims to be), the best facial recognition systems are equivalent to 1998 fingerprint matching technologies, yielding a 90% verification rate with a 1% false acceptance rate. This is a substantial improvement since the FRVT 2000, where the verification rate was 80%.

For the first time, researchers evaluated demographic factors impacting the ability to recognise faces. These results show that males are easier to identify than females, and older people are easier to recognise than younger people.

The study also found significant differences in matching abilities depending on whether the images were taken indoors or outdoors. Face recognition performance for outdoor images is only about half as good as for indoor images, where there is better control of lighting conditions.

Ten companies participated in the tests, which involved matching 121,589 images of 37,437 individuals.

The complete report is available in PDF format <a href= ‘http://www.itl.nist.gov/iad/894.03/face/face.html#FRVT2002’>here</a>.