摘要
针对当前图像中文本定位算法普遍存在定位文本精确度不高的缺点,本文提出了一种有效的图像文本定位方法(MSITE)。算法使用均值漂移方法对图像进行分割后,用区域生长的方法对分割图像进行连通域分析得到一系列可能包含文本的图像块,然后根据字符特征组合连通域并删除非字符连通域得到候选字符区域,再根据位置和属性特征进行合并,最后根据文本特征判断得到最终文本区域。试验表明该算法具有较高的准确率。
A new method of text location based on connected component is presented in this paper for current methods aren＇t accuracy in location text positions. First, Mean-shift is used to divide a image, and then an region growing method is exploited to find the spatial connectivity of pixels to form the connected regions, some character feature is used to connect some connected regions and to remove some non-character regions, then the candidate text region are connected based on the character region depended on the locations and propertys, finally the text features are used to extraction the last text regions. Experimental results show that the algorithm has high accuracy.
出处
《微计算机信息》
2009年第28期123-125,114,共4页
Control & Automation
关键词
文本定位
均值漂移
特征分析
连通分量
text location
Mean-Shift
feature analysis
connected component
