基于双分支模型的汉字字体级联分类  

Chinese Character Font Cascade Classification Based on a Dual-branch Model

作　　者：陈芯芯 石华岳 王存睿 战国栋 CHEN Xinxin;SHI Huayue;WANG Cunrui;ZHAN Guodong(School of Computer Science and Engineering,Dalian Minzu University,Dalian Liaoning 116605,China;School of Design,Dalian Minzu University,Dalian Liaoning 116605,China;Dalian Chinese Font Design Technology Innovation Center,Dalian Minzu University,Dalian Liaoning 116605,China)

机构地区：[1]大连民族大学计算机科学与工程学院,辽宁大连116605 [2]大连民族大学设计学院,辽宁大连116605 [3]大连民族大学大连市汉字计算机字库设计技术创新中心,辽宁大连116605

出　　处：《大连民族大学学报》2025年第1期66-71,共6页Journal of Dalian Minzu University

基　　金：辽宁省技术攻关计划项目(2024JH2/102600108);大连市创新基金项目(2023JJGX026)。

摘　　要：提出一种汉字字体分类方法,构建CNN和Transformer的双分支模型,利用Res Ne St和Swin Transformer作为两个特征提取分支,并在每一层中引入特征融合模块,有效捕捉汉字字体的局部和全局特征;引入级联分类,先对字体的大类进行分类,然后根据大类进行细分类。在包含104类汉字字体图像的数据集上进行实验,并与其他分类网络进行对比,本模型在汉字字体分类识别中准确率达到98.51%。结果验证了本模型在处理类间差异较小的汉字字体分类任务中的有效性,为解决汉字字体版权问题提供了有效的技术途径。This paper presents a Chinese font classification approach,builds a dual-branch model consisting of CNN and Transformer,employs ResNeSt and Swin Transformer as two feature extraction branches,and incorporates a feature fusion module in each layer.This effectively captures the local and global features of Chinese characters.Cascade classification is introduced,where the broad categories of fonts are classified first and then subcategorized based on the broad categories.Through experiments on a 104-type Chinese font image dataset and comparison with other advanced classification networks,the accuracy of this model reaches 98.51%.The results demonstrate that the model is effective in handling the task of Chinese font classification with minor differences among classes and provides an effective technical solution to issues of Chinese font copyright.

关 键 词：字体分类 双分支模型 特征融合 级联分类 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]