Handwritten Character Recognition System for Bangla Text Containing Modifiers and Overlapped Characters

This paper deals with the design and development of a Bangla offline Handwritten Character Segmentation system. Special focus of this work was on skewed text containing modifier and overlapped characters. The text pages were scanned using flatbed scanner and saved as 256 gray-level image in bmp file format. This image file was used as the input to the developed system. In order to segment the skewed lines, the text was divided into a number of vertical stripes called frame. Each frame was segmented by horizontal pixel scanning method and concatenated in proper order to form the original text line. Based on vertical pixel scanning method lines were segmented into words. In character segmentation, isolated characters were segmented by vertical scanning method but connected and overlapped characters are segmented using degenerated lower chain. The system was tested for handwritten text pages of five individual writers. The average segmentation accuracy rate of the system was about 94%.