基于Ada-boost的文字检测方法  

Text Detection Method Based on Ada-boost Learning

作　　者：李晨轩[1] 齐开悦[1] 陈凯[1] 陈剑波[1] 管海兵[2] 

机构地区：[1]上海交通大学信息安全工程学院,上海200240 [2]上海交通大学计算机科学与工程系,上海200240

出　　处：《信息安全与通信保密》2010年第4期59-61,共3页Information Security and Communications Privacy

摘　　要：提出了一种新的基于机器学习的文字检测方法。首先,在网上下载280张图片,其中包括书的封面、CD封面和电影海报,人工标记和提取其中的文字区域。其次,基于对文字区域和非文字区域的统计性差异分析,得到两大类特征集,用于构造弱分类器。然后,使用Ada-boost将上一步得到的弱分类器筛选和组织起来,得到一个二级的级联分类器。最后,通过将图片的子区域分类为文字和非文字区域,此级联分类器就能够检测出文字区域。与其他方法相比,此方法在检测单个字符、倾斜甚至竖直的文字方面有很好的鲁棒性。This paper proposes a learning-based text detection method in camera images. First, the 280 pictures of book covers, CD covers and movie posters shot with cameras are unloaded from the Internet. The text regions in them are manually labeled and extracted. Second, based on statistical analysis of the difference between text and non-text samples, the three sets of features are obtained for producing the weak classifiers. Third, Ada-boost is utilized to select and combine these weak classifiers into a two-stage cascade classifier. Finally, this two-stage cascade classifier could detect text area in images by classifying sub-regions of images as text and non-text. As compared with traditional methods, this method is more robust in detecting single characters, skewed and even vertical lines.

关 键 词：文字检测 机器学习 统计分析 Ada—boost算法 

分 类 号：TP391.43[自动化与计算机技术—计算机应用技术]