Face Recognition System And Key Algorithms Research

In today’s Internet age, e-commerce, social networking and other Web applications are increasingly popular. The related information security problems become the focus of attention, and the demand for efficient authentication technology is more and more urgent. As a highly reliable means of authentication, face recognition has been regarded to more and more people and has potentially broad application and important research value.This paper starts by the framework of face recognition, aiming to study relating algorithms in every part of the system. In this paper, the main work content is as follows:(1) Study of image pre-processing methods. An adaptive image enhancement method is analyzed and studied in this paper. Parameters in this algorithm are determined according to the global gray-level of the image, so when applied to image that has globally low and locally high gray-level, enhanced result is unsatisfactory. To improve it, local enhancement is employed on image areas selected with the sliding window. Considering the variation of illumination, illumination normalization method based on homomorphic filtering is proposed in this paper. Face images processed with this method have almost consistent illumination, so, naturally, the effect of variation of illumination is removed.(2) Study of face detection methods. Lots of false positives appear when employing face detection method based on Adaboost algorithm in complex scenario. To solve this problem, face detection combining segmentation of skin color and Adaboost algorithm are studied in this paper. Skin color verification is applied on every detected window obtained through face detection based on Adaboost algorithm, thus effectively eliminating false detections and improving face detection accuracy.(3) Study of face representation methods. To extract more effective representing information from face image, we study face representation based on multi-feature fusion. Single and simple face features including LBP and Gabor are extracted and then fused at feature level to get eventually resulting face representation. Experiment results indicated that the feature fusion method can integrate good qualities of multiple features and improve the accuracy of face recognition effectively.(4) Implementation of face recognition system. Supported by department project, a face recognition system was designed and implemented by integrating relevant algorithms proposed in this paper. This system consists of face registration, face recognition and information management module and so on, which can meet the needs of practical application.