Improving the parametric Gaussian classifier using neural networks

Abstract

The statistical approach to pattern recognition is among the early approaches applied in this field of research. This paper presents a mixed statistical parametric and neural networks approach for classifiers design. Statistical parametric techniques have the advantage of being mathematically tractable but are often non-optimal due to the need of making some assumptions about the shape of the distribution of the input data samples (e.g. being a multivariate normal distribution) and the need to estimate the distribution parameters (e.g. the mean vector and the covariance matrix) from the training data. On the other hand, neural networks classifiers are model (distribution) free. Therefore, they can be used to improve the performance of an initially given statistical classifier. Computer simulation results are given that show the efficiency of the proposed technique.