IMPROVING THE STAT STICAL PARAMETRIC GAUSSIAN CLASSIFIER USING NEURAL NETWORKS

Abstract

The statistical approach to pattern recognition is among the early approaches applied in this field of research. Parametric statistical classifiers design techniques have been extensively studied, in general, and Gaussian classifiers, in particular, due to its analytical tractability [10]. However, some assumptions inherent in the design ofthe Gaussian classifier result in suboptimal classifier [9].

Recently [33], it was demonstrated, both theoretically and experimentally, that a neural network pattern classifier generates the empirical distribution of the sample data which are used to train the network. This thesis takes advantage of this fact and improves a Gaussian classifier using an isomorphic sigma-pi neural network [9].

This study contains four chapters, and three appendices. Their contents are as follows:

Chapter 1: Presents different approaches to pattern recognition. Special attention is given to classifiers designed using the decision theoretic approach such as neural networks classifiers and traditional statistical classifiers. Both of these classifiers are discussed in more detail.