Research On Bangla Handwritten Numeral Recognition Using Directional Element Feature

Handwritten digit recognition is one branch of character recognition, and it has a great practical meaning. At present, zip codes are extensively used in letter communicating. Based on the project of Bangla Post Automatic Letter Sorting Machine manufactured by China Post Group, we focus the issue of Bangla handwritten numeral recognition in this dissertation.Handwritten numeral has ten categories only, but very high recognition precision is required in our system. In addition, the change of the handwritten numeral is very large. So it's still a tough job to deal with handwritten numeral recognition. For this, we implement this system using Director Element Feature and BP net. The system consists of image preprocessing, feature extraction, and BP net classifier.We first implement image preprocessing on Bangla handwritten numeral images including image smoothing, removing noising, binarization, character size normalization, thinning, etc. Then, DEF feature with ECP feature are employed according to the special structure of Bangla handwritten numerals. The combined feature is able to describe each Bangla numeral category and distinguish one from others. And the relatively low dimension we got greatly improves the efficiency of the classifier. At last, we obtain the best parameter for BP net by training to implement the recognition system. Experimental results show that the method is of high reliability and strong robustness in recognizing Bangla handwritten numerals and meets the practical requirements.