A Hybrid Character Segmentation Approach for Cursive Unconstrained Handwritten Historical Modi Script Documents

Abstract

Segmentation of archaic handwritten degraded Modi script text lines in the isolated characters is the key steps of the offline automated Modi script document recognition system. The improved segmentation of characters gravitated towards the higher accuracy of the recognition. The intricacy of archaic handwritten degraded Modi script documents put forwards the number of challenges in character segmentation. This paper presents a multistep technique for Modi script character segmentation based on the analysis of background pixel density of global and local horizontal zones of the text line. At the initial isolation stage two types of fragments are segmented as isolated individual characters and cluster of overlapping/ touching characters. Further, the cluster of overlapping or touching characters are segmented in isolated individual characters. The isolated, overlapping, and touching characters are segmented efficiently using the proposed Modi character segmentation technique.

Keywords: Modi Script, Character Segmentation, Density Analysis