Character recognition using Fourier descriptors and a new form of dynamic semisupervised neural network

Abstract

A new neural network is presented for pattern recognition tasks. This new network, called the Dynamic Supervised Forward-Propagation Network (DSFPN), although based upon the unsupervised Counterpropagation Network (CPN), trains using a supervised algorithm. In addition it allows unsupervised dynamic growth of the learning layer permitting unknown subclasses to be learnt. The classification capabilities of the network are tested using handwritten numerals presented as Fourier descriptors. The results are compared with those of a Back Propagation Network (BPN) and a Counter-propagation Network. This comparison shows the new network can provide as high a classification accuracy as the BPN and train in a time comparable to the CPN. On average the DSFPN trained in 1 1353 of the epochs required to train the BPN whilst producing a considerably higher accuracy than the CPN.