The Face Recognition Technology (FERET) program, which began in September 1993, was the effort that turned face recognition from a scattering of incomparable lab demos into a measurable field. It was sponsored by the US Department of Defense Counterdrug Technology Development Program Office, with Dr. P. Jonathon Phillips of the Army Research Laboratory as technical agent, and it was run through the National Institute of Standards and Technology (NIST).
FERET had three components. First, it funded algorithm development through contracts with groups at MIT, Rutgers, the University of Southern California, the University of Illinois at Chicago, and TASC. Second, and most lasting, it created a standardized database: NIST records that the FERET database contains 14,126 images of 1,199 individuals, collected across 15 sessions between August 1993 and July 1996. Third, it ran structured, independent evaluations in August 1994, March 1995, and September 1996.
The key methodological idea was independence. As the program put it, FERET “set out to establish a large database of facial images that was gathered independently from the algorithm developers.” Because every group was tested on the same images that none of them had collected, the FERET evaluations made it possible, for the first time, to fairly compare competing approaches and to track real progress over time.
That template - a common dataset plus a neutral referee running blind tests - became the standard model for biometric evaluation. It led directly to NIST’s later Face Recognition Vendor Tests, the long-running series that today measures commercial face recognition accuracy and demographic error differentials across hundreds of algorithms.