Face Recognition under Illumination Changes Using Color Fast and Adaptive Bi-Directional Empirical Mode Decomposition

Abstract

Recently, the importance of face recognition in color images has been increasingly emphasized since popular CCD cameras are distributed to various applications. However, face feature extracted from the image will distorted non-linearly by lighting variations in intensity or directions, so this changes will cause a serious performance degradation in face recognition. Many algorithms adopted by researchers to overcome the illumination problem. Most of them need multiple registered images per person or the prior knowledge of lighting conditions. According to the “common assumption” that illumination varies slowly and the face intrinsic feature (including 3D surface and reflectance) varies rapidly in local area, high frequency feature represents the face intrinsic structure. The Fast and Adaptive Bi-dimensional Empirical Mode Decomposition FABEMD has been extended for color image analysis. The proposed algorithm, based on the powerful transform for the color image named color FABEMD (CFABEMD). The color image decomposed into multi-layer high frequency images representing detail feature and low frequency images representing analogy feature. In addition a two measurements are proposed to quantify the detail feature that use to eliminate illumination variation, with these measurement weights, CFABEMD based multi-layer detail images recognition can be done under vary illumination. With PCA, the experiment results based on processed CVL database and Georgia Tech Face database show the method can get remarkable performance