Curved Segmentation Path Construction for Unconstrained Handwritten Hangul Segmentation

Abstract

Up to now, most Korean character recognition has been developed on the assumption that it was segmented into characters perfectly. It is difficult to adapt the method that is used in English, since there are spaces in a character. Because a Korean character, which we call a syllable, is composed of two or more graphemes on 2D plane, and also the characteristic of touch that occurred between character is often occurred between graphemes. In this paper, we propose the method that uses a virtual segmentation path for segmentation of unconstrained handwritten Korean characters. The main property of the proposed method is that not only the curved segmentation path which is using the space between character is used, but also the straight line segmentation path is used. The segmentation process is divided into two steps. In the first step, a straight line segmentation path of a Korean word is determined using preprocessing and vertical projection, and words that are overlapped or touching are passed to the next step. In the second step, feature points are extracted for making a number of virtual curved paths, and then the most suitable path is determined using center of gravity which is marked between three adjacent points.