Support Vector Machine Using In Face Recognition

Support vector machines, the implementation of Structural Risk Minimization (SRM) rules, have some attractive merits, such as global optimization, simple structure, generalize abilities and so on. So the development and application about SVM developed widely in recent years. As a new machine learning algorithm, SVM also has some limitations. This thesis gives two improved Proximal SVM algorithms. In One hand, it improves the proximal SVM to adapt to small sample sets. And on the other hand, the GEPSVM also is improved in using high-dimensional small sample sets and multi-class problems.Totally speaking, the thesis includes the following parts:(1) What SVM and face recognition were studied and developed in recent years show in this part. And the implementation of several standard SVMs and several classifies, such as Nearest-Neighbor classifier, Nearest Feature Line classifier and Linear/Nonlinear PSVM classifier, and some comparison results in four opening face databases also be told in this part.(2) In this part, the author presents an improved SVM algrithom which is an improving one for PSVM which changes it to adapt any kind of high-dimensional database, the multi-class classification problem also is solved in this part. The dimension reducing uses several GLDA/Kernel GLDA algrithoms.(3) In this part, the author discusses a new kind of SVM named GEPSVM in details, and also gives an improved algorithm that can solve the small sample size problem which causes the singular problem, and gives an implementation for multi-class problem at the same time.The conclusions in this thesis drawn from experiments on four opening face databases indicate the characteristics of face recognition and rules of the SVM which are important theoretical basis for their applications and fusion bio-information system. It is also a reference to SVM researchers and users.