Key Technology Research Of Face Recognition Based On Video Stream

The human face recognize is a popular and complicated problem in pattern recognition. In recent years, both the sparse representation which based on classification (SRC) and the collaborative representation which based on classification (CRC) have been successfully shown the superiority. Comparing to the conventional face recognition methods, the SRC and CRC methods have the advantages of simpleness and roubustness. The main ideas of SRC and CRC algorithms are:by building a redundant dictionary, the test sample can be expressed as a linear combination of the train samples. And then the identify label of the test sample can be obtained by the coefficient of the linear representation which contain discriminative information. But these two algorithms have some limitations:both two methods need to construct the redundant dictionary so the small sample size problem becomes one of the main problems of the algorithm. And the SRC algorithm is time-consuming because it is an optimal solution problem based on  norm constraint. Due to the l2 norm of the CRC algorithm does not have sparse property, the result of CRC is not so satisfied when the feature dimension of the samples is low.In practical applications, the problem of face recognition and monitoring of video streaming are more and more concerned by people. However, it is still difficult to recognize face from video. Occlusion, angle and lighting problems will greatly reduce the accuracy of various kinds of recognition algorithms in the theoretical study.This paper tries to solve these problems from the aspects of theory and practical application by doing in-depth analysis and research of SRC and CRC models, the main researches are as follows:(1) Do further study of the background, implementation process as well as the difference and relationship of SRC and CRC models based on compressed sensing theory.(2) Proposed a method namely sparse joint collaborative representation models of face recognition (S_CRC), which is based on SRC, CRC models and the Elastic Net linear regression method. S_CRC method use the l1 norm and l2 norm as constraint conditions making the linear representation coefficient contain more discriminating information, thus it is more conducive to classification. The feasibility of the proposed method is verified on AR and Extended Yale B face databases.(3) Found the trend of SRC, CRC and S_CRC models, summarize the difference and relationship about them.(4) Under the Visual Studio 2010 development environment, using the C++ language and vision library of OpenCV, programming MFC-based video streaming face recognition system. The main modules of this system consist of face detection, face tracking, face recognition as well as personnel information registration and image acquisition. Using SQL Server database records personal data and face image information to adapt to the larger application environments. For both the function of the experimental conditions and real-time application of monitoring environment, this system not only have the fuction of automatic face detection, tracking and recognition processing but also provide step-by-step interface processing operation on static image and video files, so it has a certain practical value.