Handwritten Character Recognition using Wavelet Transform for Feature Extraction

Abstract

Handwritten Character Recognition is one of the important area of pattern recognition. In this paper 13 sets of handwritten characters are collected from different users; features are extracted by using multilevel 2 dimensional wavelet decomposition technique. Wavelet families used are Daubechies and Reverse Biorthogonal. Wavelet decomposition is done up to three levels i.e. level 6,7 & 8.Features obtained are then train in WEKA3.6 machine learning software for different classifiers like Multilayer perceptron, K-Nearest Neighbor, Naive Byes Results obtained for different classifiers are compared with each other. It is observed that multilayer perceptron gives accuracy of 92%for 8th level of decomposition and 98% for 6th level of decomposition. Confusion matrix shows that there is confusion between characters with similar shape like O and 0, S and 5. Keywords: - Character recognition, DWT, multilayer perceptron, Decision tree neural network