基于条件生成对抗网络的书法字笔画分割

毛笔书法作为中华传统艺术的精华, 需要在新的时代背景下继续传承和发扬. 书法字是以笔画为基本单元组成的复杂图形, 如果要分析书法结构, 笔画分割是首要的步骤. 传统的笔画分割方法主要利用细化法从汉字骨架上提取特征点, 分析交叉区域的子笔画拓扑结构关系来分割笔画. 本文分析了传统笔画分割基于底层特征拆分笔画的局限性, 利用条件生成对抗网络(Conditional generative adversarial network, CGAN)的对抗学习机制直接分割笔画, 使提取笔画从先细化再分割改进为直接分割. 该方法能有效提取出精确的笔画, 得到的高层语义特征和保留完整信息的单个笔画利于后续对书法轮廓和结构的评价.

Stroke Segmentation of Calligraphy Based on Conditional Generative Adversarial Network

As the essence of Chinese traditional art, brush calligraphy needs to continue to inherit and carry forward in the new era. Calligraphy is a complex figure composed of strokes as the basic unit. If you want to analyze the structure of calligraphy, stroke segmentation is the first step. The traditional stroke segmentation method mainly uses the refinement method to extract feature points from the Chinese character skeleton, and analyzes the sub-stroke topology relationship of the intersection region to segment the strokes. This paper analyzes the limitations of traditional stroke segmentation based on the underlying feature splitting strokes, and the strokes are directly segmented by using the adversarial learning mechanism of conditional generative adversarial network (CGAN). Improve the method of extracting strokes from first refinement and then segmentation to direct segmentation. This method can effectively extract accurate strokes. The resulting high-level semantic features and individual strokes that retain complete information are helpful for the subsequent evaluation of the outline and structure of calligraphy.