An illumination invariant face recognition algorithm

Abstract

A face recognition technique, introducing an identification algorithm able to preserve facial content information independently of illumination variations, is proposed in this paper. Content information of images, stored in the face database, is kept by retaining their singular vectors while their illumination conditions are adapted according to the luminance information of the face image to be identified, provided by its singular values. To make the algorithm suitable for real time applications a fast presorting of the database is executed based on a similarity measure of block DCT coefficients. Experimental results depict the efficiency of the proposed method, showing not only a remarkable increase in the recognition rates, but also an improvement of the robustness of identification.