Two-class Linear Discriminant Analysis for Face Recognition

Abstract

In this paper, we present a novel face recognition system that uses two-class linear discriminant analysis for classification. In this approach a single M-class linear discriminant classifier is divided into M two-class linear discriminant classifiers. This formulation provides many advantages like more discrimination between classes, simpler calculation of projection vectors and easier update of the database with new individuals. We tested the proposed algorithm on the CMU PIE and Yale face databases. Significant performance improvements are observed, especially when the number of individuals to be classified increases.