Nonlinear Discriminant Features Constructed by Using Outputs of Multilayer Perceptron

This paper 1 proposes a method to extract nonlinear discriminant features from given input measurements by using outputs of multilayer Perceptron (MLP). Linear Discriminant Analysis (LDA) is one of the best known methods to construct linear features which are suitable for class discrimination. Otsu showed that LDA can be extended to nonlinear if we can estimate Bayesian a posteriori probabilities. Recently, MLP have been successfully applied to many kinds of pattern recognition problems. It is also regarded that outputs of MLP trained for pattern classi cation approximate Bayesian a posteriori probabilities. Thus we can construct nonlinear discriminant features that maximizes the discriminant criterion by using outputs of MLP as the estimates of Bayesian a posteriori probabilities.