Design And Implementation Of Face Recognition System Using PCA/ICA Based On SoC

Face recognition has been a hot research topic and widely used in many fields.Traditional face recognition algorithms are usually implemented on PC.With the improvement of the performance,especially the emergence of multi-core embedded systems based on ARM+FPGA(Field Programmable Gate Array)in recent years,the embedded systems have many advantages in portability,scalability,cost,power consumption and other aspects,which promotes the rapid development and wide application of embedded face recognition technology.The ARM or DSP(Digital Signal Processor)-based embedded platform is flexible and easy to develop,but it is slow.FPGA has strong mathematics operation ability and high speed,but FPGA's development is difficulty and workload.Regarding the issue above,Xilinx provides an all programmable system on chip solution based on ARM+FPGA.Based on the in-depth study of PCA(Principal Component Analysis)and ICA(Independent Component Analysis)feature fusion,this paper proposes a design scheme for face recognition using PCA+ICA and BP(Back Propagation)neural network,and implements on a system on chip.The main research content of this paper includes:(1)Research on Face Feature Extraction Based on PCA+ICAThe thesis researches the basic principles,advantages and disadvantages of PCA and ICA.For PCA only uses the second-order statistical information and ignores the higher-order statistical information of face image,it proposes a feature extraction method using PCA+ICA.First,PCA extracts features of the training face set and make up the eigenface space,then ICA processes the eigenface space in order to make full use of the high-order statistical information of the image.The improved recognition rate of the algorithm is 98.33%,which is 5.2% higher than the feature extraction using PCA alone.(2)Research on BP Neural Network Classifier Based on FPGAThe thesis researches the basic principles of BP Neural Network and the advantages of BP neural network classifier in nonlinear fitting and recognition rate compared with traditional nearest neighbor classifier.Starting from the network structure of BP neural network,the thesis improves the structure of its forward propagation for the characteristics of FPGA to reduce the complexity of the network structure and the difficulty of programming.(3)Implementation of Face Recognition Algorithm Based on ZYNQ AP SoCThe advantages and disadvantages of several commonly used face recognition algorithm embedded platforms are compared and analyzed.In order to overcome the shortcomings of traditional embedded platforms,the all programmable system on chip is adopted as the hardware platform for face recognition algorithms.The developing and verifying use the collaboration of software and hardware.The implementation scheme of the face recognition algorithm based on system on chip is proposed.And the FPGA of the system on chip is used to accelerate the algorithm.After acceleration,the face recognition time of the algorithm in ORL face database is less than 9 ms.