NSCT-based Adaptive Illumination Processing for Face Recognition

Abstract

A NSCT-based method is presented for adaptively extracting illumination invariants of face images, for the purpose of removing the effect of lighting variations on the algorithm for face recognition. This study begins with the use of logarithm transform to face images, and the transformed images are then decomposed into low-frequency sub-bands and high-frequency directional sub-bands by NSCT. After that, the adaptive threshold of each band is found out according to the distribution of NSCT coefficients in high-frequency sub-bands, with which the sub-bands are filtered by a compromised threshold function. Finally, inverse NSCT is applied to the filtered sub-bands, with the illumination invariant obtained. Experimental results on the two well-known face databases of Extended Yale B and CMU PIE show that the proposed algorithm can effectively eliminate the effect of illumination on face images with satisfactory recognition rates.