基于谱残差与笔画宽度变换的显著性文本检测方法  

Salient Text Detection Method Based on Spectral Residual and Stroke Width Transform

作　　者：陈霄[1] 沈洪健 李菲[1] 王希 温长吉[1] 

机构地区：[1]吉林农业大学信息技术学院,长春130118 [2]吉林农业大学科技出版部,长春130118

出　　处：《吉林大学学报（理学版）》2017年第6期1528-1532,共5页Journal of Jilin University:Science Edition

基　　金：吉林省教育厅"十三五"科学研究规划项目(批准号:2016175;2016186)

摘　　要：为有效利用Android智能终端实现对文本信息的有效识别,提出一种基于谱残差和笔画宽度变换的显著性文本特征提取方法.首先应用基于谱残差的显著性检测获取显著图;然后利用最大熵判别方法分割并建立显著性文字候选区域;最后通过笔画宽度变换算法在候选区域内提取闭合边缘等宽特征.构建基于Android智能终端和服务器的C/S架构实验平台,在MSRA-TD500文字定位数据库与实验室场景数据库中的实验结果表明,该算法可较好地获取图像中的显著性文字特征,在保证运算效率的同时算法性能有所提升.In order to effectively recognize the information of text by using Android intelligent terminals,we proposed a text feature extraction method based on spectral residual（SR）and stroke width transform（SWT）.Firstly,the salient map was obtained based on salient detection of the spectral residual.Secondly,the maximum entropy discriminant method was used to segment and establish salient text candidate regions in salient map.Finally,the SWT algorithm was used to extract characteristics of equivalent width of closed edge in the candidate region.The C/S architecture experimental platform was constructed based on Android intelligent terminal and server,and experimental results on both the MSRA-TD500 text dataset and laboratory dataset show that the algorithm can effectively obtain salient text feature in the image,and the performance of the algorithm is improved when the computing efficiency is guaranteed.

关 键 词：谱残差 笔画宽度变换 Android终端 文本特征检测 

分 类 号：TP39[自动化与计算机技术—计算机应用技术]