Orthogonal Regularized Linear Discriminant Analysis for face recognition

Abstract

In the paper the Orthogonal Regularized Linear Discriminant Analysis (ORLDA) for face recognition is proposed, which introduces the orthogonal idea to improve regularized linear discriminant analysis. The algorithm not only overcomes the singularity problem but also improves greatly the classified performance under little training samples. The effectiveness of our proposed algorithm is illustrated by Yale, YaleB, UMIST and AR face database.