Research On The Method Of Face Illumination Compensation Based On Quotient Image

Illumination, pose and expression are three very important problems in face recognition systems. Especially the changes of natural illumination cannot control by human being, and it has very unavoidable effect on face detection, face feature extraction, face dynamic tracking, and three-dimensional facial modeling, etc. This paper thoroughly analyzes the influences of illumination angle changes upon face images, and has high regard for the illumination question in face recognition. Based on Quotient Image Theory and symmetry of face structure, we have make illumination compensation of face image by half face light images. Furthermore, based on two different illumination compensation methods, we have implemented illumination compensation on PCA feature subspace and wavelet transform subspace respectively for the whole face images. The major tasks done by the paper are summarized as follows:(1) The influences of illumination upon face images and face recognition systems are analyzed. Furthermore, typical methods of face illumination compensation are summarized, and the existing problems in those methods are probed. Finally, combining with the problems of Quotient Image Theory, we give an improved method on it.(2) The calculate method of Quotient image is proposed by half face light images set, and face descending dimension illumination training set is constructed. The richness of illumination training set directly affects the estimating of the illumination condition of image recognition. In order to estimate the illumination condition of image under extreme situation, and based on the thought, which was put forward by Lee, of recognizing face image under natural illumination by using 9 typical face illumination conditions, 9-dimension illumination training set is designed. Based on this work, and combined with symmetry of face structure and location of point lamp-house, de-dimension processing of the set are realized by using PCA and wavelet transform. Furthermore, low dimension matrix in PCA subspace and wavelet transform subspace are constructed respectively. The experiment results show that it can reduce the influences of face self-shadow.(3) Two strategies of gray image illumination compensation are given, and based onwhich implement the illumination processing of color image. Adding light and reducing light are two basic thoughts in solving illumination. Based on the thought of adding light, the 9 different light images under natural illumination situations are synthesized. In addition, based on the thought of reducing light, the standard illumination image is reconstructed. Meanwhile, in PCA feature subspace and wavelet transform subspace they are realized respectively. Finally, the illumination compensation of color images has realized by separating the brightness of the color image. The statistics of the experiments shows that the recognition rate of face image has increased by 20% compared with that of the original image...