Off-line Handwritten Hindi Consonants Recognition System using Zemike Moments and Genetic Algorithm

摘要

Developing an efficient character recognition system is supposed to be a very challenging research problem. In the present work, an offline handwritten Hindi character recognition technique is proposed using Zernike moments as the descriptor of character image with a feature selection algorithm. For feature selection, use of the Genetic algorithm is proposed to reduce the length of the feature vector. The core idea of the paper is to first generate the significant Zernike complex moments and then to select the most relevant moments using the Genetic algorithm which are in turn used to classify the individual characters. The significance of low-order as well as high-order Zernike moments is also studied in recognizing the first ten consonants of Hindi script. Two resilient backpropagation classifiers are trained one for the feature vector without selection and another one for feature vector obtained after selection. The average character recognition accuracies obtained are 90% and 94.3%, respectively.