Reserch On Face Recognition Under Varying Illumination

With the development of information technology, face recognition is not only widely used in access control systems, case detection, e-commerce and other fields have also been applied to the face recognition system. Therefore, face recognition has a very high research value in today's society. Face recognition often influenced by various factors in everyday practice. Such as facial expressions, shelter materials, illumination and gestures and so on. When the face recognition system is collecting face information, the light conditions are usually uncontrolled. So how much faster and more concise removal of varying light on face recognition system has become an important research area of face recognition.This paper mainly studies the field of illumination variation in face recognition, the main research contents and innovative points are as follows:(1) This paper introduces several illumination normalization algorithm, such as:logarithmic transformation, gamma correction, histogram specification and histogram equalization. Then these types of algorithms were tested in Yale B face database, the results show that although the illumination normalization algorithm is simple and weakened the influence of light in a certain extent. If the illumination condition is very poor, only using the above conventional illumination normalization algorithm, the effect is not obvious on eliminating varying illumination.(2) This paper focuses on self-quotient image, multi-scale Retinex algorithm and gradient-face three extraction algorithm illumination invariant features. These three feature extraction algorithms are tested in the Yale B face database. The test showed that the three algorithms can get better illumination invariant features from the face images. The features mainly include the contour and surface texture details. These three algorithms have obvious inhibition to the light. However, the image is processed by MSR algorithm, and the details of the face are not completely preserved, the SQI algorithm enlarged high frequency noise in low SNR region, eighted Gaussian filter can hardly keep intact edge information and the parameter in the algorithm is complex to choose. The gradient face algorithm is not only simple to calculate and the extracting illumination invariant feature information is more complete, and the final recognition rate is better than the other two kinds of illumination invariant feature extraction algorithm.(3) In this paper, we find that the noise can seriously interfere with the gradient information, so that the recognition rate is greatly reduced. Traditional noise reduction methods will blur the image, and the less image gradient information will let the recognition rate decline. In this paper, we propose a NSCT-Gradientfaces algorithm based on the combination of the non down sampling contour wave transform (NSCT) and the gradient face extraction method. And this method has a good recognition on the noise images. The algorithm and the Gaussian filter, wavelet transform and Contourlet Transform combine face gradient algorithm test at Yale B face database, the results demonstrate that the algorithm the paper propose has better performance than the other several algorithms.