Fuzzy Support Vector Machines And It's Application On Face Recognition

Fuzzy Support Vector Machines(FSVM) is developed from Support Vector Machines(SVM) based on Fuzzy membership function. it resolves the limitation of SVM and is widely applied in the area of pattern recognition and article intelligence. it is also an important method in face recognition. Therefore, the optimization of fuzzy support vector machine algorithm, through better fuzzy support vector machine classifiers to improve the classification accuracy rate, which can be better used in face recognition. This article mainly contains three parts based on the idea:1.Kernel function is the core content in the FSVM. many characteristics of FSVM are determined by the type of kernel function being used. In general, there are two main types of kernels, named local kernels and global kernels. They have their advantages and disadvantages respectively. This paper combines the good characteristics of two or more functions,forming mixtures of kernels function。So that the mixtures of kernels have advantages of generation ability of global kernels and learning capacity of local kernel. And applied the mixtures of kernels in FSVM,The experimental results obtained from the wine data set and iris data set show the effectiveness of the proposed method.2.Similarly, the weight and bias, which is used in QP, is one of the reason which impact on classified performance of fuzzy support vector machine. Bias excessive or too small, that can directly effect the position of optimal hyperplane and the classified performance. This paper proposes a new approach to solve the bias,the optimal hyperplane of FSVM made by the new bias, and the results of experiments shows that the accuracy rate can be improved by using the proposed method according three data set.3.Kernel methods provide a disciplined to the nonlinear generalization of many linear methods. FSVM, kernel principal component analysis and kernel Fisher discriminant(KFD) are just some of the better known kernel methods. However, the advantage of a kernel method often depends critically on a proper choice of the kernel function. We propose in this paper a adaptive kernel matrix learning method based on optimizing a class separability criterion that is similar to those used by linear discriminant analysis and KFD. Kernel matrix which gained from normal kernel function mapping. According learning this matrix, that can gain optimal kernel matrix.Then replace the kernel matrix of FSVM with optimal kernel matrix, and it can form adaptive Fuzzy Support Vector Machines.