On the Integration of Neural Classifiers through Similarity Analysis of Higher Order Features

Abstract

A novel methodology is herein outlined for combining the classification decisions of different neural network classifiers. Instead of the usual approach for applying voting schemes on the decisions of their output layer neurons, the proposed methodology integrates higher order features extracted by their upper hidden layer units. More specifically, different instances (cases) of each such classifier, derived from the same training process but with different training parameters, are investigated in terms of their higher order features, through similarity analysis, in order to find out repeated and stable higher order features. Then, all such higher order features are integrated through a second stage neural network classifier having as inputs suitable similarity features of them. The herein suggested hierarchical neural system for pattern recognition shows improved classification performance in a computer vision task. The validity of this novel combination approach has been investigated when the first stage neural classifiers involved correspond to different Feature Extraction Methodologies (FEM) for shape classification. The experimental study illustrates that such an approach, integrating higher order features through similarity analysis of a committee of the same classifier instances (cases) and a second stage neural classifier, outperforms other combination methods, like voting combination schemes as well as single neural network classifiers having as inputs all FEMs derived features. In addition, it outperforms hierarchical combination methods non performing integration of cases through similarity analysis.