The FERET evaluation methodology for face-recognition algorithms

The paper addresses two prerequisites the authors identify for building reliable face-recognition systems: a large, shared database of facial images and a common testing procedure for evaluating systems. The Face Recognition Technology (FERET) program supplies both, pairing the FERET database of facial images with an established series of FERET tests. At the time of writing the database contained 14,126 images from 1,199 individuals, partitioned into a development portion and a sequestered portion held back for testing.

In September 1996 the program administered the third test in the FERET series, with three primary objectives: to assess the state of the art in face recognition, to identify promising areas for future research, and to measure algorithm performance on large databases. By standardizing both the data and the evaluation protocol, the FERET methodology provided a common basis for comparing face-recognition algorithms.