基于深度学习的汉字识别方法研究综述  

Survey of Chinese character recognition methods based on deep learning

作　　者：邓杰荣 梁森 曹昕妍 杨晓辉[2] 曹喜信[1] DENG Jierong;LIANG Sen;CAO Xinyan;YANG Xiaohui;CAO Xixin(School of Software&Microelectronics,Peking University,Beijing 100871,China;Tsinghua University,Beijing 100091,China)

机构地区：[1]北京大学软件与微电子学院,北京100871 [2]清华大学,北京100091

出　　处：《微纳电子与智能制造》2020年第3期73-81,共9页Micro/nano Electronics and Intelligent Manufacturing

基　　金：类脑视觉处理技术(YBN2018085207)项目资助。

摘　　要：中国汉字博大精深,拥有数千年的文化背景与历史积淀,具有其独特的文字结构,是世界上使用人口最多的文字。与一般的视觉元素不同,汉字包含了丰富的高层语义信息,不管在工业界还是学术界对汉字识别技术都有极大的应用需求和重要的研究意义。首先阐述了汉字识别技术的基本概念和定义,对OCR、印刷体识别、手写体汉字识别等概念进行详细说明。在简要介绍了传统汉字识别方法后,详细分析了当前基于深度学习的汉字识别方法的研究历程、主要方法和关键性技术。最后总结了当前汉字识别技术在深度学习应用中存在的一些问题及未来的发展趋势。Chinese characters are vast and profound,have thousand years of cultural background and historical accumulation,have their unique text structure,and are the most populous text in the world.Different from general visual elements,Chinese characters contain rich high-level semantic information.No matter in the industrial field or academia,Chinese characters'recognition technology has great application requirement and important research significance.This article first elaborates the basic concepts and definitions of Chinese character recognition technology,details the concepts such as OCR,print recognition,handwritten Chinese character recognition.After introducing the traditional Chinese character recognition method briefly,the research history,main methods and key technologies of current Chinese character recognition method based on deep learning are analyzed in detail.Finally,it summarizes some existing problems and future development trends of current Chinese character recognition technology in deep learning applications.

关 键 词：汉字识别 深度学习 OCR 卷积神经网络 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]