Thai Character Segmentation in Handwriting Images using Four Directional Depth First Search

Abstract

© 2019 IEEE. One of the key processes for converting handwriting images into digital texts is the character segmentation. It is very challenge especially for the case of segmenting the hand-writing due to intra-variations of various writing styles and overlapping of characters between consecutive characters. This paper works on Thai characters in handwriting images. Thai characters consist of different types of consonants, tones and vowels, which are written in different manners. This paper proposes the 4 directional depth first search based approach for segmenting individual characters in both vertical and horizontal cutting aspects. The vertical cut is applied to segment each text column, while the horizontal cut is applied to segment individual characters. Then, the erosion with two structuring elements is used to split overlapped consecutive characters that may be remained after the main segmentation process. The proposed method is validated with 11,949 Thai characters in handwriting images. It achieves up to 90.76 % of the successful segmentation.