Illumination normalization for robust face recognition using edge-preserving filtering

A new illumination normalization method based on the Retinex illumination model for face recognition is proposed. Total variation under L2 constraint (TV-L2) model and wiener filtering are used to estimate illumination, respectively. Their capability of edge-preserving can impair the notorious halo effect. Illumination normalized face image can be obtained by subtracting the luminance image from the original image in logarithm domain. For validating the effectiveness of the proposed method, experiments have been conducted on the representative Yale B and CMU-PIE face databases. The Experimental results show that the presented method can weaken the effect of uneven illumination effectively and improve face recognition rate.