A New Two-Phase Method of Face Recognition Based on Image Matrix

Abstract

This paper proposes a two-phase algorithm of image projection discriminant analysis. The new discriminant method is composed of feature extraction by on maximum margin criterion (MMC) and Fisher discriminant analysis (FDA). The algorithm includes two stages: firstly, the feature extraction based on maximum margin criterion (MMC) is employed to condense the dimension of image matrix; Then Fisher discriminant analysis (FDA) is applied to reduce dimension of condensed image matrices. This novel method based on image matrix is called 2DMMCplu2DFDA in the paper. Different from the previous linear discriminant analysis method for face recognition where FDA or PCA is based on image vector, 2DMMCplus2DFDA is to exploit image matrices to directly construct the between-class scatter matrix, within-class scatter matrix and total-class scatter matrix. The experimental results on ORL face databases indicate that the proposed method is more efficient and stable than 2DPCA, 2DMMC and 2DPCAplus2DFDA with higher recognition rate.