Recognition-based segmentation of on-line cursive Korean characters

The Korean language has a large set of characters and many of them are very similar in shape. Therefore, it is very difficult to separate graphemes from a handwritten character without a contextual knowledge. In this paper, we propose a recognition-based stroke segmentation technique using grapheme as a recognition unit. Our method uses a time delay neural network recognition engine and a graph-algorithmic postprocessor based on the Korean grapheme composition rule and Viterbi algorithm. We experimented the proposed method on freely handwritten characters and the result obtained are given