Illumination Normalization for Face Recognition under Extreme Lighting Conditions

Abstract

An effective illumination normalization method based on human visual system is presented for extreme lighting face recognition. One contribution is that illumination normalization based on retinal modeling is mainly executed on low frequency band considering lighting conditions, the other is the introduction of discrete wavelet transform into human visual modeling for illumination normalization. The proposed method not only gets better illumination normalized result, but also preserves more image details. Both of them are very important for face recognition under complex lighting conditions. Experimental results on extended Yale B face databases demonstrate that our method is effective for dealing with variable lighting, especially for extreme lighting variation situation.