Eliminating Variation of Face Images Using Face Symmetry

Abstract

In many face recognition tasks the pose and illumination conditions of the probe and gallery images are different. In most cases, gallery and probe images may only be two or three images, each captured from a different pose and under different illumination condition. The variation of the images caused by illumination and pose is acute. We propose a method, which reduce the illumination influence by the input image itself. The revised images can improve the face recognition performance. The basic idea of the method is to reconfigure the probe image in the subspace of the illumination space, so as to reduce the variation of the images caused by illumination. We utilized the fact that the face is almost bilateral symmetry. We also propose a method to estimate the direction of the face, which utilizes the fact too.¹

The work is supported by the National Natural Science Foundation of China (60175004).