摘要
本文对自然场景文本检测问题及其方法的研究进展进行了综述.首先,论述了自然场景文本的特点、自然场景文本检测技术的研究背景、现状以及主要技术路线.其次,从传统文本检测以及深度学习文本检测的视角出发,梳理、分析并比较了各类自然场景文本检测方法的优缺点,并介绍了端对端文本识别技术.再次,论述了自然场景文本检测技术所面临的挑战,探讨了相应的解决方案.最后,本文列举了测试基准数据集、评估方法,将最具代表性的自然场景文本检测方法的性能进行了比较,本文还展望了本领域的发展趋势.
In this paper,the research progress of natural scene text detection problems and methods are reviewed.Firstly, the characteristics of natural scene text are introduced,and the text detection technology research background,status and the main technical route are illustrated respectively.Secondly,from the perspective of traditional text detection and deep learning text detection,the merits and demerits of various methods are analyzed,and the technology of end-to-end text recognition is introduced.Then,the challenges of the natural scene text detection technology and the corresponding solutions are discussed.Finally,the benchmark datasets are enumerated,the evaluation methods and the performances of the most representative approaches are fundamentally compared.Furthermore,potential application and development trend in this field are summarized.
作者
王润民
桑农
丁丁
陈杰
叶齐祥
高常鑫
刘丽
WANG Run-Min;SANG Nong;DING Ding;CHEN Jie;YE Qi-Xiang;GAO Chang-Xin;LIU Li(School of Physics and Information Science,Hunan Normal University,Changsha 410081,China;Department of Information System and Management,National University of Defense Technology,Changsha 410000,China;School of Automation,Huazhong University of Science and Technology,Wuhan 430074,China;Teaching and Research Support Center, National University of Defense Technology,Changsha 410000, China;Department of Electrical and Information Engineering,University of Oulu,Oulu,FI-90014,Finland;School of Electronics,Electrical and Communication Engineering,Univer sity of Chinese Academy of Sciences,Beijing 101408,China)
出处
《自动化学报》
EI
CSCD
北大核心
2018年第12期2113-2141,共29页
Acta Automatica Sinica
基金
国家自然科学基金(61502164)
湖南省自然科学基金(2016JJ3090)
湖南省教育厅优秀青年项目(16B155)
中国博士后科学基金(2015T81130
2014M562569)资助~~
