Connected Component Analysis Based Two Zone Approach for Bangla Character Segmentation

In this paper a complete and robust segmentation process for printed Bangla script is presented. This phase is important for language like Bangla in which resources are not easily available. In Bangla OCR system, one of the main challenges is character segmentation. Proper segmentation increases the accuracy of recognition process. Segmentation is difficult as there are more than 300 basic, modified and compound characters in the script and they are topologically connected. Some characters are overlapped. We have used connected component analysis based two zone approach that can addresses all the problems associated with character segmentation of Bangla script with less effort. In our system we have tried to identify the matra line properly. Then using this information we have divided the script into two zones (upper and lower zone). Connected components from each part are then collected by overlooking matra line. Using some heuristics segmented character parts are organized for proper recognition. Proposed method solves the problems of character segmentation in Bangla script and thereby increases the accuracy of segmentation phase that leads to a greater accuracy of overall optical character recognition (OCR) system.