A high precision printed character recognition method for Tamil script

In this paper, we describe an efficient printed Tamil character recognition method developed for an omni font case using HOG (Histogram of Oriented Gradients) features. A Back propagated Neural Networks classifier is used at the original classification stage, which is followed by a secondary FLDA (Fischer Linear Discriminant Analysis) classifier to overcome the misclassifications by the BPN network. An algorithmic fusion approach is used to improve the accuracy of the primary classifier. A comparison of the performance of the proposed classification technique with different features is also provided. There is no benchmark database to conduct studies on the printed Tamil character recognition. So, a database of 3670 samples of varying sizes with 35 font types has been created for our research.