Improved principal component analysis and linear regression classification for face recognition

Abstract

In this paper, an improved principal component analysis (IPCA) is presented for face feature representation. IPCA is mainly designed to extract the useful information from original face images through reducing the dimension of feature vectors. Linear regression classification (LRC) algorithm is employed to treat the face recognition as a linear regression issue. LRC uses the least-square method to decide the class label with the minimum reconstruction error. Experiments are conducted on the Yale B, CMU_PIE and JAFFE databases. The proposed IPCA algorithm and LRC algorithm achieve better recognition results than that of state-of-the-art algorithms.