Illumination Normalization for Robust Face Recognition Based on Nonsubsampled Contourlet Transform

Abstract

In this paper, we propose a novel illumination normalization method for face recognition under varying lighting conditions. In the proposed method, the logarithm transform is firstly applied to facial image under various lighting conditions, which transfers illumination model from multiplicative model to additive one. Then the adaptive normal shrink filter based on the nonsubsampled contourlet transform (NSCT) is used to obtain illumination invariant, which is applied to face recognition. Experimental results on Yale B and CMU PIE databases show that the proposed algorithm can eliminate the effect of illumination for face recognition.