CSLDA and LDA fusion based face recognition

Abstract

Face recognition has great demands and become one of the most important research area of pattern recognition but there are several issues involved in it. Unsupervised statistical methods i.e. PCA, LDA, ICA are the most popular algorithms in face recognition that finds the set of basis images and represents faces as linear combination of those images. This paper presents a novel layered face recognition method based on CSLDA and LDA. The basic aim is to decrease FAR by reducing the face dataset to very small size through layered linear discriminant analysis. Although the computational complexity at the time of recognition is much higher than conventional PCA and LDA because weights are computed for small subspace at time of recognition but it provide a good results especially for large dataset. CSLDA of LDA is insensitive to large dataset and also small sample size and it provided 84% accuracy on Banca face database. The proposed approach is also applicable on other applications and recognition methods i.e. PCA, KDA, DLDA etc.