Enhanced discriminant linear regression classification for face recognition

Abstract

Linear Discriminant regression classification (L-DRC) embeds the fisher criterion into the linear regression classification (LRC) and can achieve more robust classification performance for face recognition. In this paper, we propose an enhanced discriminant linear regression classification (EDLRC) algorithm to further improve the discriminant power of LDRC. When calculating the between-class reconstruction error (BCRE), only those classes that are more easily to be misclassified into are considered. After maximizing the ratio of BCRE and within-class reconstruction error (WCRE), the obtained projection matrix in EDLRC is more effective than the projection matrix in LDRC, which is verified by extensive experiments.