CHARACTER RECOGNITION FOR BI-LINGUAL MIXED-TYPE CHARACTERS USING ARTIFICIAL NEURAL NETWORK

Rekha Singh

Abstract: Artificial Intelligence has evolved to a great extent since its inception and is spreading wide day to day. Today, nearly all aspects of human life have a sound presence of some sort of application of artificial intelligence. A key role in this is of Artificial Neural network, a very vital branch of Artificial Intelligence. A vast majority of day by day applications have their evolvement from Artificial Neural Network, especially applications developed for those tasks which were earlier believed to be best performed by only human beings. This potential of ANNs lies in its degree of simplicity and generality; bringing the difference down in between, Human mind capability and machines. This paper basically deals with recognition of isolated machine print as well as hand-written characters using artificial neural networks. The important aspect of this paper which makes it different from similar works in this field is that I have approached this problem with bilingual character set. I have approached to the recent trend where people are at home with use of English as well as Hindi language, not only in speaking but also in writing. Due to this, my focus has been to device network with almost equal capability of recognizing characters in both scripts; Devnagri and English. Besides this, text content faced by us in day to day is a mixture of type written and hand written. So the characters are either computer generated or by human handwriting. For this purpose data collection, feature extraction and neural network designs have been achieved. The training phase has successfully been completed with 790 samples of handwritten samples computer generated font styles. The overall performance of the system has been achieved more than 95%, quite significant in case of human handwriting.

Keywords: Artificial Neural Network; Character, Recognition; Feed-forward Back-propagation Network

DOI: https://doi.org/10.15623/ijret.2014.0322017