Introduction of a new notion into RBF networks: the shape concept

In order to realize an industrial character recognition system, we have developed a new notion in the RBF network system. For this supervised classification problem, we have shown that the large training set generates a complex and slow in recall mode architecture for most of neural networks classifiers. Thus, for the chosen classifier, the RBF network, the size of the architecture and, then, the processing time in running mode are proportional to the number of models in the training set. We have opted to adapt the RBF model in order to limit its size for a large training set and to reduce the processing time. We have modified the RBF network in order to introduce the concept of shape. The introduction of the shape concept has led us to modify the input function of the RBF kernel-function neurons, keeping a Gaussian activation function