An improved retinal modeling for illumination face recognition

Yong Cheng, Liangbao Jiao, Zuoyong Li, Xuehong Cao

Abstract: Illumination variation is one of the most important challenges for robust face recognition system under real environment. It attracts more and more attention in face recognition field. In this paper, an improved retinal modeling is proposed to alleviate the adverse effect of lighting variation on face recognition. There are two main contributions. One is that it develops a new scheme to calculate appropriate adaptation factor through maximum filtering and illumination classification. The factor is quite crucial for illumination normalization by modeling the retinal information processing mechanism. The other is that an adaptive truncation based on the median statistics is used for contour enhancement. The proposed method can preserve image details, while achieves good illumination normalization results. Experimental results on the Extended Yale B face databases show that the new method achieves high recognition rates, and is quite effective in varying lighting condition, especially in difficult lighting situation.