Multiresolution technique to handwritten English character recognition using learning rule and Euclidean distance metric

The present paper deals with the problem of handwritten character recognition of English character. This paper presents a novel method of handwriting character recognition which exploits a compression capability of discrete wavelet transform to enhance the accuracy of recognition at the pixel level, the learning capability of artificial neural network and computational capability of Euclidean distance metric. The problem of handwritten character recognition has been tackled with multiresolution technique using discrete wavelet transform and learning rule through the artificial neural network. Recognition accuracy is improved by Euclidean distance metric along with recognition score in case of misclassification. Features of the handwritten character images are extracted by discrete wavelet transform used with appropriate level of multiresolution. Handwritten characters are classified into 26 pattern classes based on appropriate properties i.e. shape. During preprocessing each character is captured within a rectangular box and then resized to a threshold size. Weight matrix of each class is computed using the learning rule of artificial neural network, and then the unknown input pattern vector is fused with the weight matrices of all the classes to generate the recognition scores. Maximum score corresponds to the recognized input character. Learning rule provides a good recognition accuracy of 88.46%. In case of misclassification, the Euclidean distance metric improves the recognition accuracy to 92.31% and then its product with recognition score further improves the recognition accuracy to 99.23%. The proposed method provides such good recognition accuracy for handwritten characters even with fewer data samples.