一種手寫中文字的筆劃擷取法

Abstract

本論文提出一種新的手寫中文字筆劃擷取法。本方法先將文字分割成若干 區域或稱為節點 (Node), 並用圖形(Picture Graph)來表示節點間的關係 。 一個節點若有一個以上的祖先 (Ancestor) 或子孫 (descendant) 便 被認為是交差節點 (Junction Node)。若由交差節點的祖先向子孫長(或 反方向)可得到此二節點間的重複比例 (Overlapped Ratio)。根據任一筆 劃寬度應幾乎保持一定值及筆劃中某一段的方向性並無太大變化,這兩項 觀察,屬於同一筆劃的兩節點有著相當高的重複比例。如果重複比例大於 事前設定的臨界值 (Threshold)則此二節點便視為同一筆劃。而藉此,手 寫中文字的筆劃可被成功擷取出來。 A New stroke extraction method of handwritten Chinese characters is proposed in this paper. The approach firstly encodes an input character as a run-length form. Based on this form, the character is partioned into several regions. A picture graph is made to represent the relations among regions. The picture graph reflects a fact that a stroke might be cutted into several regions by other strokes. A region with more than one region above it or below it would be regarded as a junction of strokes. Two important facts are that the width of a stroke is almost near constant and the orientation of part of stroke does not vary a lot. Based on these facts, the proposed extractor will grow upper regions to lower regions of a junction region or grow lower regions to upper regions and then count the overlapped ratio between growing and grown nodes. If the overlapped ratio is greater than a predefined threshold the growing and grown regions will be merged and regarded as a same stroke. The nodes belonging to the same stroke will have high overlapped ratio due to the facts described above. No pixel- based thinning process is used in this paper. Some experimental results are available to show the effectiveness of the proposed approach.