A Robust Modified Character Segmentation Approach for the Handwritten Archaic Modi Documents

Character segmentation is an imperative step of intelligent handwritten archaic Modi document recognition system. Numbers of challenges like inconsistent and non-uniform handwritten characters, broken or degraded characters etc. are faced in the process of character segmentation due to age, cursive and stylish writing nature of Modi script. This paper presents a modified preliminary segmentation step for segmentation of Modi intact characters and overlapping/touching characters cluster to overcome problem of under/bad segmentation. Here, a zone based three steps Modi script character segmentation approach is presented. At preliminary step, column wise background and foreground pixels are scrutinized to determine leading segmentation of text line. Modi text line is segmented in two types of segments as isolated characters and cluster of overlapping/touching characters. Local zoning based method and column wise background pixel density exploration is used to segment overlapping and touching characters clusters respectively. Performance of the proposed method and comparative analysis is verified using MODI-HHDoc Modi document dataset. Successful Modi character segmentation rate is achieved as 87.70%. As compared to previous hybrid background pixel density based technique, bad segmentation rate is reduced from 0.8 to 0.2%. Comparative analysis shows that proposed modified Modi character segmentation technique is more efficient to state-of-art benchmarking techniques.