An end-to-end text spotter with text relation networks
作者机构：Institute of Information EngineeringChinese Academy of SciencesBeijingChina School of Cyber SecurityUniversity of Chinese Academy of ScienceBeijingChina Centre for Cyber Security Research and InnovationDeakin UniversityGeelongAustralia
出 版 物：《Cybersecurity》 (网络空间安全科学与技术（英文）)
年 卷 期：2021年第4卷第1期
页 面：91-103页
核心收录：
学科分类：081203[工学-计算机应用技术] 08[工学] 0835[工学-软件工程] 0812[工学-计算机科学与技术（可授工学、理学学位）]
基 金：国家自然科学基金
主 题：Scene text spotting Graph convolutional network Visual reasoning
摘 要：Reading text in images automatically has become an attractive research topic in computer ***,end-to-end spotting of scene text has attracted significant research attention,and relatively ideal accuracy has been achieved on several ***,most of the existing works overlooked the semantic connection between the scene text instances,and had limitations in situations such as occlusion,blurring,and unseen characters,which result in some semantic information lost in the text *** relevance between texts generally lies in the scene *** the perspective of cognitive psychology,humans often combine the nearby easy-to-recognize texts to infer the unidentifiable *** this paper,we propose a novel graph-based method for intermediate semantic features enhancement,called Text Relation ***,we model the co-occurrence relationship of scene texts as a *** nodes in the graph represent the text instances in a scene image,and the corresponding semantic features are defined as representations of the *** relative positions between text instances are measured as the weights of edges in the established ***,a convolution operation is performed on the graph to aggregate semantic information and enhance the intermediate features corresponding to text *** evaluate the proposed method through comprehensive experiments on several mainstream benchmarks,and get highly competitive *** example,on the SCUT-CTW1500,our method surpasses the previous top works by 2.1%on the word spotting task.
维普期刊数据库