AN ANALYSIS OF HANDWRITTEN CHARACTER RECOGNITION USING NEURAL NETWORK AND STATISTICAL METHOD

Abstract

Handwritten Character recognition is a critical area of research. In this paper pattern classification method for recognizing handwritten numerals in the database using Neural Network and Statistical method is presented. MATLAB has been used as a programming Tool. The data file consists of 15,000 handwritten numerals labelled into ten classes from 0 to 9 each of size 30 X 30 pixels. 600 dimensional depictions of each numeral image is cumbersome, hence few features are extracted from the numerals that best represent numerals and give desirable accuracy. The present study is focused on two methods for pattern classification. One is neural networks and the other is statistical approach. But more concentration is on neural networks, the statistical method used here is just for comparison.The most widely used Multilayer Perceptron network with one hidden layer is selected for classification in neural networks, whereas Bayesian theory is selected for Statistical methods. Six features are extracted from the primary data set by calculating the skewness, kurtosis, mean, standard deviation, variance, and entropy for the X and Y coordinates of every image. The vector of obtained patterns is divided into training dataset and testing dataset. First dataset is used to coach the classifier using a back propagation training algorithm, whereas the second one to test the performance of the classifier. The results obtained from the neural network are then compared with the Bayesian classifier. It has been found that accuracy to classify the numerals is achieved more using neural networks as compared to statistical methods.