On occlusion robust, illumination and facial expression invariant face recognition

Abstract

In face recognition, the presence of occlusions, illumination variations and facial expressions are known to greatly reduce recognition performance. nPCA and nLDA approaches based on classical PCA and LDA has been proposed in this paper to cope with these difficulties and improve the recognition performance. The proposed scheme employs the use of histogram equalization and mean-and-variance normalization. Tests have been carried out on the JAFFE database in order to compare performance with classical PCA and LDA.