The Research Of Face Recognition System Based On Subspace

Personal identification based on biological characteristic has been widely used in allareas of our lives. Compared with other technologies of personal identification, facerecognition, a distance and fast identification technology that could be applied without user’scomplicity, has become the latest trend and research focus in relevant areas presently. Facerecognition, being involved in digital image processing, pattern recognition, computertechnology and psychology, and other science fields, has the important scientific researchvalue and widespread application prospect.A systematic research for face recognition had been done in this paper. DetailEnhancement PCA for image processing was proposed and a face recognition system wasdesigned. The major research results were shown as follows:Firstly, Adaboost algorithm and classifier training process have been investigated deeply.And face detection and localization with Adaboost algorithm has been realized on personalcomputer.Secondly, preprocessing algorithms of face recognition, based on image processing,including set normalization, gray normalization and binary have been researched.Thirdly, sub-space face recognition algorithm is the mainly research in this paper,including PCA and LDA. Based on PCA algorithm, Detail Enhancement PCA, which fusedthe local standard deviation of image enhancement processing and histogram equalizationmethod, was proposed that can improve dynamic range of image and strengthen local details.Additionally, when tested directly in ORL and YALE face image database, DetailEnhancement algorithm was proved to improve the recognition rate and have certainrobustness against light and expression change. In order to solve small sample problem in theLDA algorithm, 2DPCA+LDA, the hybrid algorithm of face recognition, were designed basedon 2DPCA and fisherface. The experimental results proved that Detail Enhancement 2DPCA+LDA improve recognition rate.Finally, a human face recognition system, including face training and recognitionmodules, was designed, and also tested to show the functions and recognition results. Testshave shown that this system can realize face recognition in a simple environment.