The paper develops an autoassociative neural network (AANN) based online handwritten character recognition method for Tamil scripts. The coordinate positions traced during the pen movement in the digital tablet are preprocessed to remove noise, normalize the characters to a uniform height, and rescale the pen positions to result in uniformly spaced x and y positions. The rescaled x and y positions are individually provided as feature vectors to an AANN classifier and trained using the back propagation algorithm. The network training results in the adjusted weights that minimize the error between the input and output and captures the distribution of feature vectors in the input space effectively to create separate x and y models for 156 Tamil characters. The number of rescaled points is varied in order to experiment with various AANN structures and determine the best recognition rate. The confidence scores individually obtained from the x and y models are concatenated to give rise to the weighted xy model through a weighted sum rule. The experimental results reveal a higher recognition rate of 89.74 % for a weighting factor of 0.3 when applied to the scores of 96 rescaled points. The strength of the proposed approach lies in the computational simplicity and the consistency of the results claim its use in real world applications. © 2013 Springer.
CITATION STYLE
Sigappi, A., & Palanivel, S. (2013). AANN-based online handwritten Tamil character recognition. In Lecture Notes in Electrical Engineering (Vol. 188 LNEE, pp. 35–42). https://doi.org/10.1007/978-81-322-1035-1_4
Mendeley helps you to discover research relevant for your work.