Research On Face Recognition Based On Local Features

With the development of hardware technology, computer performance has been improved dramatically and made the real-time face recognition possible. Compared with other biometric features recognition technologies, the face recognition technology has many advantages, such as high degree of automation, friendly interface and non-intrusive characteristics. Until now, under the controllable conditions of pose, illumination and the cooperation of the users, face recognition technology has been relatively mature. However, when these conditions are uncontrollable, the performance of the face recognition system will be deteriorated sharply. The face recognition under complex conditions is just conducting under these uncontrollable conditions. Meanwhile, real-time is one of the difficulties to apply this technology to the reality. Considering these difficulties, this paper focuses on the study of local feature based face recognition to find an appropriate algorithm which can adapt to the technical requirements of the face recognition under complex conditions. Thus, the main content of this research can be summarized as follows.Firstly, this paper expounds the components and fundamentals of face recognition system briefly. Besides, it also analyzes the difficulties of face recognition.Secondly, the algorithm of invariant feature based has invariant advantages in rotation, illumination, scaling, facial expression, and dimension in the application of face recognition. In addition, this paper proposes an improved matching method of SURF feature on the basis of existing literature. What’s more, this method makes multi-pose face recognition have better robustness after the even division of the face image and giving different weight to different cell depending on the contribution that cell similarity makes to global similarity and variation of face pose.Thirdly, this paper systematically tests and verifies the performance of the algorithm of SURF on the PIE problems of face recognition and make SURF compare to SIFT in ability of handling those problems. In addition, real-time improves obviously under the conditions of similar recognition rate and the algorithm of SIFT. In the mean time, this paper also compares the improved matching method with many existing algorithms and finds that the algorithm in this paper has advantages in multi-pose face recognition.Lastly, this paper apply a method of the WLGBPHS to solve the problem that the face images got from the image are vulnerable to the illumination and partial occlusion in non-ideal environments. The experimental results show that the method of WLGBPHS has robustness for illumination and partial occlusion, and can improve recognition rate and real-time of the face recognition system...