An Efficient Method of Nonlinear Feature Extraction Based on SVM

For nonlinear feature extraction, a new feature extraction method based on kernel maximum margin criterion (KMMC) is presented in this paper, i.e., an algorithm of statistically uncorrelated optimal discriminant vectors in kernel feature space is proposed in the paper. The proposed method has more powerful capability to eliminate the statistical correlation between features and improve efficiency of feature extraction. Our experimental results show that the new method is better than original KMMC and kernel principal component analysis (KPCA) in terms of efficiency and stability about feature extraction on Olivetti Research Laboratory (ORL) face database by leave-one-out method