Research In Face Recognition Based On Dempster-shafer Evidence K Nearest Neighbor And Design Of System

With the electronic and bio-information technology developing rapidly,the technology of pattern recognition based on biometrics such as face recognition with potentially extensive use in more and more areas has drawn attention from all the world,for its concurrency,no touching with human faces,friendly interaction and convenience..Face recognition is still one of the current research hotspots,and there are great economic benefits in the research about it.Firstly,the evidence theory is introduced into the K-nearest neighbor algorithm to improve the performance of the face recognition for its ability of multisource information fusion to process uncertain and inaccurate information in this dissertation.In this improved algorithm,the basic probability assigments(BPA)are determined by using the feature distrances between the unknown sample and its K nearest neighbors in each class of the face set,and then the BPA in each class are conbined.Finally the combined results in each class are fused,and the unknown sample can be recognized by the fusion result and the classification rule presented in the dissertation.Experiments on face databases show that the evidence K-nearest neighbor algorithm can improve the accuracy of the face recognition effectively,compared with the K-nearest neighbor algorithm.Secondly,we design and implement a face recognition system based on evidence K-nearest neighbor for face recognition(divided into administrator and normal user mode),in which the server and client of the system are designed,the message flow is designed and packaged by Protocol Buffer,the face database is stored in MySQL.The system obtains the images from the camera and completes the face recognition in the server.Again,the server can provide face recognition service for as many users as possible simultaneously by the architecture of master-slave reactor and thread pool.The use of local preferential search speeds up and improves real-time performance of the face recognition.Several detectors are trained to obtain more information to judge whether a face exists,which improve the performance of face detection under occlusion.In the end,the function and performance of the system are tested in detail.many samples including the interference of various postures,occlusion,and illumination are designed to test the performance of face detection.We adopts the automated technique to test the the performance of concurrency and throughput of the server,the accuracy and the time-spent of the recognition of the system.The result shows that the system meets the requirements and has some practical value.