Research And Implementation Of Face Verification In Video

With the development of Society and Information Technology, identity recognition technology grows every day. RFID (Radio Frequency Identification) card and second generation identity card, which both are based on radio frequency techniques, have been applied to Identification field widely recently. And Identification techniques based on fingerprints, voice and human face have been applied gradually. But most Identification techniques are based on single feature recognition and they have defects in validity and security.This paper attempt to do research work in face detection, face recognition and RFID technique, design and implement a video oriented, automatic identification system based on RFID and face recognition technology. These works utilize a lot of devices and technique research productions of E-Learning lab in Shanghai Jiaotong university.Face detection and face recognition have important application value in a lot of fields. This paper manly applies this technique to video stream analyzing. The aim is to develop a video stream oriented face verification system combined with RFID techniques. First,Adaboost algorithm is applied to cut out face pictures from video stream. Then, in order to remove lighting effect and supply convenience to following face recognition , face pictures cut out from video stream are normalized by rotation and size normalization. This paper analyses different kinds of feature extraction algorithm and face recognition algorithm and brings up a modified LDA algorithm. Taking the consideration of both target human faces and non target human faces, this paper imports"competitive negative samples"to adjust verdict curve more precisely, reduce the false verification rate and improve the stability of algorithm. At last, this paper brings up an integrated, video stream oriented and RFID combined automatic face recognition system. This system is very convenient for extension and easy to substitute algorithms which make a lot convenience for comparison of different algorithms.The structure of this paper is as follow. The first chapter is introduction which introduces basic conception and application of face detection and face recognition technology. The second chapter summarizes current face detection and face recognition algorithms. The third chapter introduces some prevail face detection algorithms, analyses the Adaboost algorithm in detail and develops a face detection software module based on the research products and the requirement of our system. The forth chapter introduces some prevail face recognition algorithms, analyses modified LDA algorithm, imports"competitive negative samples"and implements a software module for face extraction, training and verification based on the requirement of our system. The Fifth chapter introduces the architecture of the video stream oriented identification system mentioned above and describes the implementation of every sub modules in detail. The sixth chapter makes a conclusion to the whole paper and points out the future work.