Performance evaluation methodology for face recognition algorithms

We present two fundamental performance evaluation methodologies for face recognition algorithms. Our experiments include (1) the development of an evaluation methodology based on an identification and verification model and (2) the investigation of design decisions for a principal component analysis (PCA) based face recognition system. Throughout the series of experiments, we present a robust and comprehensive evaluation methodology for face recognition algorithms that allows researchers to identify the relative strengths and weaknesses of their algorithms and that points out the directions for future research.; Two critical performance characteristics of face recognition algorithms are the identification and verification performance. We report performance results based on the identification and verification model for various face recognition algorithms. We identify the state of the art by direct quantitative assessment of different approaches. The results that we report are for images taken (1) on the same day, (2) on different days, (3) at least one year apart, and (4) under different lighting conditions.; PCA-based algorithms form the basis of numerous algorithms in the face recognition literature. PCA is a statistical technique and its incorporation into a face recognition system requires numerous design decisions. We explicitly state the design decisions by implementation of a generic modular PCA-based face recognition system. We make a comprehensive analysis of the different implementations for each module, as these affect the variations in performance.