Design Of Eyes Location Algorithm And Software Porting For Face Recognition System

As one of the biometrical technologies, face recognition technology has got more andmore attention for its noncontactive, active and user-friendly. Face recognition technologyis used in various application prospects for government, army, bank, social welfare andsecurity system, electronic commerce, security system etc.As one of the most important procedures of face recognition system, eyes locationdevotes a lot to face recognition. After analyzing the research background and mainmethods of face detection and recognition technology, this thesis proposed an eyeslocation algorithm based on SVM(Support Vector Machine) and Hough circle detection.SVM performs well on non-linear classifying and high-dimension pattern recognition andhas got high accuracy on solving multi-category classification. Hough detection is notsensitive to noise of image and edge discontinuous point, so it can be used to detect theiris circles of eyes effectively. The DSP(Digtal Signal Processor) kernel ofTMS320DM6446which is produced by TEXAS INSTRUMENT has high-speedprocessing capability and abundant multimedia human-computer interaction interfaces. Inorder to overcome the disadvantages of face recognition system based on PC platform,such as high cost, high power consumption and not easy to place and carry on, weimplemented the software porting for the face recognition system from personal computerto DSP and optimized the algorithm according to the properties of DSP kernel ofDM6446.The result shows that, the proposed algorithm has got a high accuracy and is robust tosituations like polarized light and cover of decorations. After porting, the face recognitionsystem running on DSP has got the same accuracy as it is running on personal computer,and the time cost decreased a lot after optimizing.