基于颜色散布分析的自然场景文本定位  

Text Location in Natural Scenes Based on Color Distribution Analysis

作　　者：周慧灿[1] 刘琼[1] 王耀南[2] 

机构地区：[1]湖南文理学院计算机科学与技术学院,常德415000 [2]湖南大学电气与信息工程学院,长沙410082

出　　处：《计算机工程》2010年第8期197-199,202,共4页Computer Engineering

基　　金：湖南省科技计划基金资助项目"自然场景下的文本定位与提取方法研究"(2008GK3131)

摘　　要：针对现有的文本定位方法存在阈值选择困难、计算窗口尺寸敏感、文本特征与非文本特征交错难以区分等问题,提出一种文本定位方法。利用局部颜色散布分析,突出文字和背景的差异,使阈值选择范围更宽松。通过空间颜色散布分析,基于文本和背景颜色的空间分布实现文本的精确定位和分割,避开文本特征分类问题。实验结果表明,该方法适应性较好,准确性较高。There are some roadblocks in existing methods for text location.It＇s hard to choose a robust threshold for discrimination.The size of computing-window is very sensitive.Because of overlapping of characteristics between texts and non-texts,it is difficult to distinguish them.Aiming at these problem,a new method for precise text location is put forward.The new method enhances the local different between texts and backgrounds by using analysis of local color distribution,so that threshold selection becomes more easily.By an analysis of spatial color distribution,precise text location and segmentation are made.In that course,spatial color distribution of texts and backgrounds is adopted to avoid the problem of classification on features of texts.Experimental result indicates that,the new method has better flexibility and high accuracy.

关 键 词：颜色散布分析 阈值选择 文本定位 文本区域合并与筛选 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]