联合边缘增强的MSER自然场景文本检测  

Text Detection in Natural Scene Image with Joint Edge Enhanced MSER

作　　者：田清越 高志荣[2] 熊承义[1] 陈少波[1] 

机构地区：[1]中南民族大学电子信息工程学院,武汉430074 [2]中南民族大学计算机科学学院,武汉430074

出　　处：《小型微型计算机系统》2017年第11期2604-2609,共6页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61471400)资助;中南民族大学校基金项目(YCZW15118)资助

摘　　要：针对图像模糊及低对比度造成最大极值稳定区域(MSER)文本检测性能下降的问题,提出一种新的结合边缘增强的M SER文本检测方法.通过采用加权引导滤波与HOG特征结合进行边缘增强,以更好突出图像边缘增强文本信息,并结合采用Quick shift文本颜色聚类技术,有效提升了M SER文本检测的召回率;基于文本与非文本的笔画宽度差异,采用文本骨架的拓扑特征减少连通区域标记的冗余,有效提升了MSER文本检测的准确率.基于ICDAR2003数据库的实验结果验证了本文方法的有效性.Aiming at the problem that the image blur and low contrast causes the decrease of the Maximally Stable Extremal Regions（ MSER） text detection performance,a new MSER text detection method based on edge enhancement was proposed. We use the method which is based on the weighted guided image filtering and HOG features combined with edge enhancement to highlight the image edge and enhance text information,and combine to use the Quick shift text color clustering technology,which effectively improves the recall rate of MSER text detection; Based on the difference of text and non-text＇s stroke w idth,we use the topological features of text skeleton to reduce the redundant of the connected component labeling,which effectively improves the accuracy of MSER text detection. Experimental results based on ICDAR2003 database verify the effectiveness of this proposed method.

关 键 词：文本检测 最大稳定极值区域 边缘增强 文本骨架 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]