System design and assessment methodology for face recognition algorithms

Abstract

For biometric person authentication, system design and asessment methodology form essential part of the entire process. We design a performance evaluation methodology for face recognition system based on identification and verification model. To validate our model, we designed a projection-based face recognition system which requires numerous design decisions. We explicitly state the design decisions by introducing a generic modular face recognition system. We explore various implementations for preprocessing, representation and recognition modules. Our experiment includes major factors for system design and assessment: (1) changing the illumination normalization preprocessing; (2) varying the number of features in the representation; and (3) changing the similarity measure in recognition process. We perform experiments and present results for identification and verification scenarios. © Springer-Verlag Berlin Heidelberg 2004.