• 中国计算机学会会刊
  • 中国科技核心期刊
  • 中文核心期刊

计算机工程与科学

• 人工智能与数据挖掘 • 上一篇    下一篇

基于汉字笔画与结构的特征字库构造及优化

宋春晓,黄峰,靳松清,白晓东,仇宏斌,姜杰,李艺   

  1. (南京师范大学教育科学学院,江苏 南京210024)
  • 收稿日期:2018-06-05 修回日期:2018-09-27 出版日期:2019-05-25 发布日期:2019-05-25
  • 基金资助:

    国家社会科学基金“十三五”规划2016年度教育学课题(BCA160052)

Construction and optimization of characteristic fonts
based on Chinese character strokes and structures
#br#  

SONG Chunxiao,HUANG Feng,JIN Songqing,BAI Xiaodong,QIU Hongbin,JIANG Jie,LI Yi   

  1. (School of Education Science,Nanjing Normal University,Nanjing 210024,China)
  • Received:2018-06-05 Revised:2018-09-27 Online:2019-05-25 Published:2019-05-25

摘要:

以GB2312标准的6 763个汉字为例,首先以笔画骨架点集形式制作楷体字库标准模板;而后以同种形式采集书写者的45个手写汉字,从中提取笔画和部件等字素,计算基本书写特征并将其赋予标准字库模板,从而初步构造出该书写者的基本特征字库;接着利用机器学习的SVM和KNN方法,针对45个例字,对不同书写者例字集的结构特征进行计算,将所得结构特征赋予上述基本特征字库,最终得到书写者的个性特征字库,此即谓“优化”。最后进行的主观判断实验说明,优化后的字库有更高的接受度,本方法有望大大降低个性化字库的制作难度,降低书写者输入压力,缩短字库制作时间,节约字库制作成本。

关键词:

Abstract:

Taking 6763 characters of GB2312 standard as examples, we firstly make the standard templates of regular scripts and collect the handwritings of 45 individuals both in the form of skeleton point sets, from which strokes and components elements are extracted. Then the basic writing features are calculated and assigned to the standard font library templates, thus the basic feature font library of each writer is constructed. Thirdly, we use the SVM and KNN machine learning methods to calculate the structural features of the 45 samples of Chinese characters. The obtained structural features are given to the above basic feature font library of each writer, based on which, the personalized feature library of Chinese characters is constructed. This is called "optimization". Finally, the subjective judgment experiment shows that the optimized fonts have higher acceptability. The proposed method can greatly reduce the difficulty of making personalized font library, relieve the input pressure of the writers, shorten the time of making font library, and save the cost of making font library.

Key words: