Recent developments in artificial neural network based character recognition: a performance study

Artificial neutral networks (ANN) provide a robust computational paradigm for character recognition. The character classifier needs to have the capability to separate arbitrarily shaped regions in the pattern space. The recent development in such ANN-based classifiers and their learning methods are reviewed. Such classifiers are based on multilayer (hidden layer) ANN or higher-order correlation ANN and use backprojection learning. The performance of some ANN-based classifiers is evaluated and their relative performance is compared. The performance evaluation is based on factors such as recognition accuracy and reliability of classification fault tolerance to misregistration and spatial quantizations, computational costs in training, and classification.<
>