Character Segmentation of Devnagari Script in Printed Document Images using Projection Profiles

Abstract

An efficient segmentation module plays an important role in the complete OCR system as errors in the segmentation module hampers the recognition rate of OCR systems. Compared to basic characters, segmentation of modified and conjunct characters is a difficult task because of the presence of modifiers and half characters. In this paper, a novel technique based on projection profile, which also utilizes different inherent features possessed by these characters for finding an accurate segmentation path is proposed. The proposed system accurately segments basic as well as modified and conjunct characters with segmentation accuracy ranging from 91.84% to 99.11%.