Handwritten Gurumukhi Character Recognition Using Convolution Neural Network

Handwritten Character Recognition (HCR) is one of the challenging processes. Automatic recognition of handwritten characters is a difficult task. In this paper, we have presented a scheme for offline handwritten Gurmukhi character recognition based on CNN classifier. The system first prepares a skeleton of the character, so that feature information about the character is extracted. CNN based approach has been used to classify a character based on the three features like Zoning, Horizontal Peak Extent and Diagonal. We have taken the samples of offline handwritten Gurmukhi characters from 70 different writers. We have experimented partition strategy for selecting the training and testing patterns. We have used in all 2450 images of Gurmukhi characters for the purpose of training and testing. We have used Zoning, Diagonal and Horizontal Peak Extent feature extraction techniques in order to find the feature sets for a given character. The proposed system achieves a maximum recognition accuracy of 92.08% with 90% training data and 10% testing data using Zoning based features and CNN Classifier.