基于连通域的图像中场景文本定位  

Connected component based approach for scene text localization in images

作　　者：姚金良[1] 赵先章[1] 曾隽芳[1] 杨一平[1] 

机构地区：[1]中国科学院自动化研究所,北京100080

出　　处：《高技术通讯》2007年第6期612-617,共6页Chinese High Technology Letters

基　　金：863计划（2005AA420050）资助项目.

摘　　要：提出了一种基于连通域的自动定位图像中场景文本的方法。该方法充分利用了场景文本的两类特征——字符特征和文本区域特征,同时对一些字符特征进行组合,组合得到的新字符特征能够对字符的大小、字体等有很好的不变性。该方法利用级联弱分类器将所有的特征组合到一个框架中,提高了处理速度。实验结果显示,该方法对字符的大小、颜色、语言等具有很好的鲁棒性,并具有较高的召回率。A robust connected component based method was presented for automatic locating text in scene images. The proposed method utilized two kinds of text features the feature of character and the feature of text region. The new combined features of character are invariant with respect to character size, font, etc. The features were put into a cascade of classifiers which can improve the speed of the algorithm. The experiments were carried out to test the performance of the method, and it was shown that the proposed approach was robust with respect to character size, color, language, and had a high recall rate for test images.

关 键 词：连通域分析 场景文本 文本定位 Niblack 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]