A nonlinear subspace method for pattern recognition using a nonlinear PCA

Linear subspace method based on principal component analysis has been applied for pattern recognition of high-dimensional data. However, the linear subspace method can not represent nonlinear characteristics of the data-distribution efficiently. In order to overcome this problem, some nonlinear subspace methods have been proposed. In this paper, we propose a novel nonlinear subspace method for pattern recognition using multi-layered perceptrons which can hierarchically construct a nonlinear subspace from the data-distribution. We introduce the optimization scheme of the subspace taking into account the classification. The proposed method achieves high classi-fication accuracy by the optimization scheme and provides the finite subspace to avoid the crossover of the subspaces. We examine its effectiveness through some experiments.