Fpga-based Local Binary Pattern (lbp) Face Recognition Research

With the rapid development of Biological identification technology, face recognition has become the hot research topic in Biological identification field. It is studied widely in many fields and subjects, such as computer technology, electronic technology, applied mathematics, automation, visualization, virtual reality, image processing, and pattern recognition. At the same time, face recognition has been used widely in aerospace, weather, criminal detecting, exit-entry administration, airport inspection, railway station and so on.However, majority of face recognition systems are realized by software. Facing mass image data and more complex algorithm, the processing speed of them is very slow,and make them difficult to meet requirements of real-time processing Therefore, how to improve the processing speed has become the key subject in the face recognition research.Adopting Anaconda card which is special image processing card and produces by Dalsa Corporation of Canada, This paper can extract face feature by local binary pattern(LBP) algorithm. With the help of the AccelDSP synthesis tool, using the pure hardware FPGA approach can achieve real-time extracting face feature.The main content of this paper as follows:(1) This paper introduces the research background of face recognition, and descripes the related knowledge of recognition technology,such as research contents development history, existing problems, development tendency in the future,and main application field.(2) This paper introduces the related theory of local binary pattern (LBP) including the principle of the algorithm,algorithm characteristics, facial feature representation, the improved LBP algorithm. At the same time, this paper descripes the design tools of face recognition system based on the FPGA,such as design flow, coding style and comprehensive results of the AccelDSP synthesis tool. This paper introduces structure, performance parameters, image acquisition and transmission case of Anaconda card. In addition,Sapera LT class library and based class library knowledge is introduced.(3) With the help of AccelDSP synthesis tool,this paper verifies and analyzes the floating-point model, generates and verifies sentinel model and register-based model, integrates register-based model, realizes and verifies gate level model, finally completed that the LBP extracts the facial features based on FPGA.(4) Anaconda card is used to implement hardware design. This paper introduces the knowledge of hardware design, such as image bus interface, configuration interface, and firmware design. In addition, software design includes FPGA image data source setting, PC image source setting, FPGA image data output settings. After finishing the design of hardware and software, the PC will transport the face image into FPGA, and then, the PC will get facial feature which is return from FPGA to PC and make human face classified according to the features.(5)This paper has designed a face recognition system based on FPGA, and completed the face recognition experiment. Firstly, achieving the face feature extraction by write a Synthesizable local binary pattern (LBP) algorithm based on the Matlab. Secondly, it uses the AccelDSP synthesis tools to convert effectively the floating point format of Synthesizable Matlab code to hardware description language. Thirdly, the face feature extraction module can be integrated into the FPGA of Anaconda image processing card by the Xilinx ISE. Finally, under the VC++6.0development environment, using the Sapera LT class library which was provided by Dalsa Company achieve hardware control and Software operating, and using the nearest neighbor classifier has designed a real time face recognition system.Through the design and implementation of the real-time face recognition system; this paper can further verify the feasibility of this scheme from a technical point of view. So far, the development and the design of face recognition system have been completed, and the result of experiment has achieved the expected effect. It not only improves the recognition rate of system, but also has far exceeded the face recognition system based on software in processing speed. It provides a good reference value for real time face recognition system further research.