Overview

P. Jonathon Phillips, Patrick Grother, and Ross Micheals

National Institute of Standards and Technology, Gaithersburg, MD 20899, USA. jonathon@nist.gov, pgrother@nist.gov, rossm@nist.gov

Automatic face recognition has a long history of independent evaluations with the three FERET evaluations [13, 14, 15] and the two Face Recognition Vendor Tests (FRVT) [2, 11]. These five evaluations have provided a basis for measuring progress in face recognition, determining the most promising approaches and identifying future research directions.

The heart of designing and conducting evaluations is the experimental protocol. The protocol states how an evaluation is to be conducted and how the results are to be computed. In this chapter we concentrate on describing the FERET and FRVT 2002 protocols. ^[1] The FRVT 2002 evaluation protocol is based in the FERET evaluation protocols. The FRVT 2002 protocol is designed for biometrie evaluations in general, not just for evaluating face recognition algorithms.

The FRVT 2002 protocol was designed to allow us to compute a wide range of performance statistics. This includes the standard performance tasks of open-set and closed-set identification as well as verification. It also allows resampling techniques, similarity score normalization, measuring the variability of performance statistics, and covariate analysis [1, 5, 6, 9].

1 Standard Performance Measures

In face recognition and biometrics, performance is reported on three standard tasks: verification and open-set and closed-set identification. Each task has its...