A Novel Feature Extraction Method

AbstractIn contrast with the support vector machine (SVM) algorithm, the minimum class variance support vector machine (MCVSVM) classification algorithm takes into consideration both the samples in the boundaries and the distribution of the classes and gives a robust solution. In this paper, following the idea of the maximum margin discriminant analysis (MMDA) algorithm which is based on the SVM algorithm, we extend the MCVSVM algorithm to reduce the dimensionality of the sample space and propose a novel feature extraction method. We discuss both the linear case and the nonlinear case of the proposed method in the paper. The experimental results demonstrate that the proposed method yields competitive results compared with MMDA