基于支持向量机的汉字字库生成  

Chinese Character Font Generation Based on SVR

作　　者：乔琪 岳继光[1] 吴继伟[1] QIAO Qi;YUE Ji-guang;WU Ji-wei(College of Electronic and Information Engineering,Tongji University,Shanghai 201804)

机构地区：[1]同济大学电子与信息工程学院,上海201804

出　　处：《新一代信息技术》2019年第1期1-10,共10页New Generation of Information Technology

基　　金：上海市科委科技攻关项目:汉字字库计算机智能制作系统研发(11dz1505202)。

摘　　要：在传统的汉字字库制作过程,每个汉字从设计制作到后期调整的全过程都需要造字专家参与,人工成本高,制作效率低。本文针对结构较稳定的印刷字体,提出了一种基于组合造字法的汉字字库生成模型,结合汉字的结构特征和机器学习的方法,对小规模样本字库中的汉字按结构进行拆分,通过支持向量机回归方法学习得到同一结构的不同汉字由构件到整字之间的通用关系模型,利用这一模型将汉字构件加工成其他非样本汉字,最终得到大规模字库。在基础系统架构上,加入样本抽取和特征约简,将模型简化。实验结果表明,针对印刷体的汉字,通过对规模在全字库6.5%-13.5%的小样本字库进行训练,得到大规模汉字字库生成模型,保证中心和重心的误差均在2%以内。In the traditional Chinese character font production process,each Chinese character needs to participate in the whole process from production to adjustment.The labor cost is high and the production efficiency is low.With the development of the digital information age,the demand for different fonts is gradually increasing.The defects of the traditional Chinese character library production method are more and more obvious.In order to make up for these shortcomings,this paper proposes a Chinese character font generation model based on combined word-making for the more stable printed fonts.Combining the structural features of Chinese characters with the machine learning method,the Chinese characters in the small-scale sample fonts are structured according to the structure.The splitting is performed to obtain the general relationship model between different Chinese characters of the same structure by using the support vector machine regression method.The Chinese character component is processed into other non-sample Chinese characters by using this model,and finally a large-scale font is obtained.The experimental results show that for the Chinese characters of the printed body,the model can train a small sample font with a scale of 1000-2000 to obtain a large-scale Chinese character font that basically meets the structural feature requirements and aesthetic requirements.

关 键 词：汉字字库 智能化系统 组合造字法 支持向量机 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]