Research On Face Recognition Methods And DSP Implementation

In recent years, face recognition technology is a very hot topic in the field ofpattern recognition and machine vision. Due to the use of human facial features, facerecognition has convenience and no invasive, which other biometric methods can notmatch advantage. And it is widely used in the user login authentication, pay foridentification, security monitoring and many other fields. Face feature extraction isvulnerable to the influence of many factors such as illumination changes andexpression, and the existing methods are still not perfect. So the research space islarge. Face recognition methods based on subspace are fancied by the majority ofscholars, especially the ones based on linear subspace. Because it can extract the faceof essential features and has a strong ability of face description and a great dimensionreduction effect. The purpose of this paper is to study on the effective method for facerecognition and design a facial recognition system on DSP which has practicalapplication value. For the algorithm, we focus on studying the linear subspacemethod. The paper’s main works are as follows:(1) I study the existing face recognition algorithm, analyze the facial feature fromthe perspective of the global information and local information, and establish thebasic ideas of the extracted features which must be able to hierarchy descript the facefrom global to local. And we also analyze the algorithm based on linear subspace andbased on LBP detailedly.(2) After discussing the LDA and2DLDA method, the paper analyzes the defectsof2DLDA from the mathematical point of view and proposes a face recognitionmethod based on ULBP and2DLDA (ULBP-2DLDA). The method can hierarchicallykeep effective information of face, and get the processing space in the same timewhich can approximate the best performance of2DLDA. Experimental results showthat this method significantly improves the performance of2DLDA, and hasrobustness for changes of illumination and expression.(3) The paper has finished the design and implement of a simple and effective face recognition system. And build up a face recognition system on VS2010. UseOPENCV to develop video capture and display module. Design the processingmodule by integrates the face recognition based on ULBP-2DLDA and the facedetection algorithm based on Haar features and Adaboost cascade classifier.(4) On TDS642EVM development platform, I build a framework of videoprocessing system base on RF5on DSP. Then transplant the face detection and facerecognition algorithm to the TDS642EVM, and optimize the system according to theresources on TMS320DM642DSP chip. Finally release a rapid and accurateautomatic face recognition system on DSP, which can achieve the real-time standardon industry basically.