An Approach for Recognizing Modi Lipi using Ostu's Binarization Algorithm and Kohenen Neural Network

Abstract

Character recognition is simple for humans, but it is quite complex to build a software that can recognize characters. A cursive type Indian origin script known as Modi script is used to write various language like Hindi, Gujarati, Kannada, Persian, Tamil and Telugu. We intend to develop such a character recognition system capable to recognize printed and handwritten Modi characters at an electronic speed by scanning the documents. It is a challenging issue to develop a practical Cursive Character Recognition System (CCRS) which can maintain a high recognition accuracy without concerning the quality of input documents. We have developed Modi Script Character Recognizer System (MSCR) using Otsu's Binarization algorithm and Kohonen neural network method. We have considered 22 different characters of Modi script including vowels and consonants to train the system by using Kohonen neural network. The sample data set (training images) of the characters is maintained by taking handwritten samples from different peoples. These images are used to provide training to the system. The obtained results prove the effectiveness of the proposed recognition technique. We have got lower recognition rates of those characters which are similar in shape and structure. Overall, an acceptable character recognition rate, of 72.6%, has been achieved in the case of handwritten characters.