Off-line Odia handwritten numeral recognition using neural network: A comparative analysis

Character recognition is one of the most interesting and challenging research areas in the field of image processing. The recognition rate of handwritten character is still limited due to the presence of large variety of shape, scale and format in hand written characters. A sophisticated handwritten character recognition system demands a better feature extraction technique that would take care of such variety of handwriting. The work proposed in this paper is an attempt to develop recognizer for Odia handwritten numeral digits based on Binarization and discrete cosine transform (DCT) scheme. The recognizer puts emphasis on exploiting the inherit characteristics of Odia numeral images. The system first employs the techniques like thinning, foreground and background noise removal, cropping and size normalization etc., to preprocess the character images. Binarization and DCT technique is employed separately to extract the features from the images. Subsequently, these feature vectors are sent to the neural network classifier. Extensive simulations show that the results are very promising over a standard dataset and the recognition rate for Binarization and DCT are 80.2% and 90%, respectively.