Decision boundary feature extraction for nonparametric classification

Feature extraction has long been an important topic in pattern recognition. Although many authors have studied feature extraction for parametric classifiers, relatively few feature extraction algorithms are available for non-parametric classifiers. In this paper we propose a new feature extraction algorithm based on decision boundaries for nonparametric classifiers. We note that feature extraction for pattern recognition is equivalent to retaining "discriminantly informative features" and a discriminantly informative feature is related to the decision boundary. Since non-parametric classifiers do not define decision boundaries in analytic form, the decision boundary and normal vectors must be estimated numerically. We propose a procedure to extract discriminantly informative features based on a decision boundary for non-parametric classification. Experiments show that the proposed algorithm finds effective features for the non-parametric classifier with Parzen density estimation.