二阶段手写汉字骨架提取优化  

Two-stage Handwritten Chinese Character Skeleton Extraction Optimization

作　　者：陈泓汉 王涛[2] 熊显航 CHEN Hong-han;WANG Tao;XIONG Xian-hang(School of Artificial Intelligence,South China Normal University,Foshan 528200,China;School of Computer Science,South China Normal University,Guangzhou 510631,China)

机构地区：[1]华南师范大学人工智能学院,广东佛山528200 [2]华南师范大学计算机学院,广东广州510631

出　　处：《计算机技术与发展》2023年第7期41-46,共6页Computer Technology and Development

基　　金：国家自然科学基金项目(61402185);广州市民生科技攻关计划民生科技攻关专题项目(201903010103)。

摘　　要：骨架提取是研究手写汉字的重要先决步骤,基于细化算法提取手写汉字骨架存在笔画交叉处产生不同程度畸变以及毛刺等问题,提出一种适用于细化算法的二阶段手写汉字骨架提取优化算法。使用细化算法得到待修复的汉字骨架,通过二阶段检测算法识别汉字骨架中笔画交错分叉区域以及确定畸变待修复范围,并依据分叉区域类型,采用相应的修复策略。实验结果表明,该二阶段检测法能够有效地检测出笔画分叉区域,准确率达到96.89%,且分叉区域归类算法可很好地识别出畸变区域类型。经过修复后的汉字骨架能够正确地反映笔画间的原有拓扑结构,使得最终优化骨架更能突出手写汉字特征,验证了算法的有效性。Skeleton extraction is an important steps for the study of handwritten Chinese characters.There are some problems in extracting the skeleton of handwritten Chinese characters based on the refinement algorithm,such as distortion and burr at the cross of strokes.In order to solve the problem,we propose a two-stage handwritten Chinese character skeleton extraction optimization algorithm.A thinning algorithm is used to obtain the original skeleton.Afterward,identiying the stroke intersection area in the skeleton and determining the distortion range to be repaired by two-stage detection algorithm.According to the type of fork area,the corresponding repair strategy is adopted.The experimental results show that the two-stage detection method can not only effectively detect the stroke bifurcation area with the accuracy rate of 96.89%,but also well identify the type of distorted region.The optimized skeleton can better highlight the features of handwritten Chinese characters.The repaired skeleton of Chinese characters can accurately reflect the original topological structure between strokes,so that the final optimized skeleton can better highlight the features of handwritten Chinese characters,and verify the effectiveness of the algorithm.

关 键 词：手写汉字 细化算法 区域畸变 分叉点合并 畸变修复 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构] TN957.52[自动化与计算机技术—计算机科学与技术]