Research on Face Recognition Algorithm Based on D-S Evidence Theory and Local Domain Pattern

Abstract

Local Binary Pattern is a kind of description of the texture within the scope of gray level, but under the influence of illumination and noise, the performance of classification decline rapidly. Therefore, a local feature extraction method is proposed, that is, Local Neighborhood Patterns (LNP). First of all, the image is divided into some local block. Then, treat every pixel and the pixel within the scope of neighborhood as a vector, calculate the distance of each local regional center vector and other pixels vector. Then take the ring extraction features in each local block, it means, using the center of the coded local image as the center of the different radius circle, from the center of the external circular in a certain sequence (clockwise or counterclockwise) extraction characteristics. Make the characteristics of the local image connect one by one as a whole image characteristic vector. At last use the nearest neighbor classifier. This paper also combines data fusion with the LNP, using D-S evidence theory for decision fusion. First of all, it used the gaussian filtering illumination pretreatment method to reduce the influence of the extreme imaging conditions on the face image. Secondly, the image convolution with sobel operator, then get horizontal and vertical edge image. Thirdly, extraction feature vector with LNP and calculate the distance between the test sample and all the classes, by means of the constructor function to realize the conversion of the Euclidean distance and the objective evidence. In the end, to make optimal decision Using D-S evidence theory to objective evidence for fusion. The result in the Extended Yale B database shown that this method can not only get high recognition rate, but also can effectively improve the robustness of illumination, posture, facial expression change.