Feature extraction for face recognition using recursive Bayesian linear discriminant

Abstract

In this paper, we present two linear discriminant analysis algorithms (LDA), namely, recursive Bayesian linear discriminant I (or RBLD-I) and recursive Bayesian linear discriminant II (or RBLD-II), for the problem of face recognition. The favorable contribution of these two LDA algorithms is that they extract discriminative features with criterion functions directly based on minimum probability of classification error, or the Bayes error. The effectiveness of the two RBLD's are tested by application to two types of face recognition tasks: identity recognition and facial expression recognition. Experimental results show that the two RBLD's achieve superior classification performance over their fellow algorithm, recursive Fisher Linear Discriminant (or RFLD), on Yale, ORL and Jaffe face databases.