Study On Face Recognition With One Training Sample Based On Principal Component Analysis

Face Recognition,which began in later 1960's,obtained a rapid development in recent 40 years,made a great progress as well as achieved wide application. With the development and application of Face Recognition going deep,however, a problem called face recognition using a single training sample arose.This problem has brought great challenges to Face Recognition,and solving it could not only widen the application scope of Face Recognition but also be helpful to the solution of the small sample problem in pattern recognition.Therefore,it has attracted more and more attention.This paper conducted positive and effective researches on face recognition using a single training sample.The main work of this paper can be summarized as follows:(1) This paper summarized the current study situations both domestic and overseas for face recognition using a single training sample,analyzed the strengths and shortcomings of various methods,illustrated the challenges it faces, and predicted the future direction for it.(2) The methods PCA,KPCA and SPCA aiming at face recognition using a single training sample are implemented.What's more,the influences of the parameters of SPCA on recognition results are discussed,and the advantages and disadvantages of SPCA analyzed in this paper.(3) The unsupervised learning algorithm(2D)~2PCA is introduced to face recognition using a single training sample.More important,the algorithm is improved and a new algorithm called weighted(2D)~2PCA(W(2D)~2PCA) is presented in this paper.Subsequently,the influence of the weight on recognition results is discussed,the optimal weight explored,and the reason that W(2D)~2PCA is superior to(2D)~2PCA analyzed in this paper.(4) A method called modular weighted(2D)~2PCA(MW(2D)~2PCA) is also proposed to effectively extract the local feature of the face by combining image-blocking and W(2D)~2PCA.The influences of the ways of image-blocking are discussed,and the advantages and disadvantages of image-blocking analyzed in this paper.(5) At last,this paper designed a face recognition system using a single training sample,tested and compared the recognition results of the method proposed in the paper with those of PCA,KPCA,SPCA and others by conducting experiments on face database ORL.Experiments show that the method proposed in this paper can achieve good recognition results.