Performance Analysis of Handwritten Marathi Character Recognition with RBF, Cascade, Elman and Feed Forward Neural Networks

Abstract

Character recognition of handwritten Marathi curve scripts is one of the most challenging areas of research in neural networks due to high variability in writing styles. Marathi characters have shirolekhas and spines. This seriously affects many of the performance recognition parameters and much more.In this paper, we are performing the performance analysis of RBF neural network, Cascade Neural network, Elman Neural network and Feed forward neural network for the character recognition of handwritten Marathi curve scripts. For the experiment, we have taken in to account the six samples each of 48 Marathi characters. For every sampled character, the �Edge detection and dilation method of Feature extraction�with a set of image pre-processing operations have been performed. Here to study and analyze the performance of these four neural networks, firstly we have created the network, trained the network, simulated the network and plotted the regression plots. It has been analyzed that RBF neural networks has a high regression value as compared to the rest of the methods for the training set.