自然场景下中文文本定位关键技术的研究  

RESEARCH ON KEY TECHNIQUES OF CHINESE TEXT POSITIONING IN NATURAL SCENES

作　　者：万燕[1] 王晓华 卢达 Wan Yan,Wang Xiaohua,Lu Da(School of Computer Science and Technology, Donghua University, Shanghai 201620, China)

机构地区：[1]东华大学计算机科学与技术学院,上海201620

出　　处：《计算机应用与软件》2018年第7期243-249,共7页Computer Applications and Software

基　　金：国家自然科学基金项目(61472075)

摘　　要：针对自然场景图像的中文文本定位问题,提出一种基于汉字特征的文本定位方法。采用MSER(Maximally Stable Extremal Regions)算法提取候选文本区域;通过数学形态学运算连接汉字分离的笔画。依据汉字的特点,设计启发式规则并结合汉字的结构特征过滤非文本区域,实现初步定位;在初步定位的文本区域内作笔画宽度变换,再根据文本的笔画特征进一步过滤非文本区域。最后,通过任意方向文本行合并算法,合并定位好的文本区域,从而实现最终中文文本定位。实验证明,该方法对自然场景图像的中文文本定位取得了良好的效果。Aiming at the problem of Chinese text positioning of natural scene images,a text localization method based on Chinese characters was proposed. First,candidate text regions were extracted using the MSER（ Maximally Stable Exclude Regions） algorithm. Then,the strokes separated from the Chinese characters were connected by mathematical morphological operations. According to the characteristics of Chinese characters,heuristic rules were designed and the non-text areas were filtered by combining the structural features of Chinese characters to achieve initial positioning. The stroke width was changed in the area of the initially positioned text. Then the non-text area was further filtered according to the stroke characteristics of the text. Finally,by combining the arbitrary line text merging algorithm,the localized text region was merged to achieve the final Chinese text positioning. Experiments showed that this method achieved good results for Chinese text positioning of natural scene images.

关 键 词：自然场景 MSER算法 结构特征 笔画宽度变换 中文文本定位 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]