基于分层块过滤和笔划特征的场景文字提取方法  

SCENE TEXT EXTRACTION METHOD BASED ON HIERARCHICAL BLOCK FILTERING AND STROKE FEATURES

作　　者：柏宏飞[1] 金城[1] 

机构地区：[1]复旦大学计算机科学技术学院,上海200433

出　　处：《计算机应用与软件》2010年第5期60-63,共4页Computer Applications and Software

基　　金：国家科技支撑计划课题(2007BAH09B03);上海市科委课题(07dz15008;08dz1500109)

摘　　要：场景文字包含了重要的场景图像的语义信息。因此将场景图像中出现的文字抽取出来,将会对场景图像的内容分析、检索和浏览提供有益的帮助。提出的场景文字提取方法,是在边缘检测的基础上,使用分层块过滤的方法在不同尺度上过滤背景,产生场景文字区域,然后对聚合出来的文字区域根据笔划颜色和笔划宽度方面的特征进行二值化分割得到二值化文字图像,这些二值化后的文字区域图像可以作为OCR引擎的输入进行识别,从而达到提取场景图像语义信息的目的。分层块过滤的方法能较好地过滤背景聚合产生文字区域,利用文字的笔划特征也能有效地分割出文字笔划像素。实验结果也证明了方法的有效性。Scene text contains important semantic information of scene images.So it will be helpful for content analysis,browsing and retrieval of the scene image when the emerging text information is extracted from it.The scene text extraction method proposed in this paper is in such a way that it adopts hierarchical block filtering method to generate scene text regions first by filtering the background on different scales based on edge detecting,after that,the aggregated text regions will be executed the binarized segmentation according to stroke features of the colour and width of the strokes to acquire binarized text image,and these binarized text region image can be authenticated as the inputs of OCR engine so as to achieve the goal of extracting the semantic information of the scene images.This hierarchical block filtering method related in the paper can preferably filter complex background to generate aggregated text regions,and the segmentation of text stroke pixels can be effectively achieved by using the stroke feature of the text.Experimental results also demonstrate that this method is effective.

关 键 词：场景文字 边缘提取 分层块过滤 笔划特征 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]