基于图像区域的交互式文本图像阈值分割算法  

Interactive Document Images Thresholding Segmentation Algorithm Based on Image Regions

作　　者：龙建武[1,2] 申铉京[1,2] 陈海鹏[1,2] 

机构地区：[1]吉林大学计算机科学与技术学院,长春130012 [2]符号计算与知识工程教育部重点实验室(吉林大学),长春130012

出　　处：《计算机研究与发展》2012年第7期1420-1431,共12页Journal of Computer Research and Development

基　　金：国家自然科学基金项目(60773098);吉林省科技发展计划基金项目(20080317);吉林大学研究生创新基金项目(20121104)

摘　　要：针对现有局部阈值分割算法因参数过多带来的参数选择问题以及在分割结果中块与块之间不连续性问题,利用用户提供的先验知识或经验,提出了一种基于灰度图像区域的交互式文本图像阈值分割算法.该方法首先粗略地将图像进行分块;利用标准差作为衡量图像块含有信息量(背景信息与目标信息)多少这一度量,接着按标准差大小对所有图像块进行排序;然后由用户输入交互式信息将所有图像块分为3个集合:仅含背景或仅含目标的图像块、含有少量背景或者是含有少量目标的图像块以及背景和目标分布比较均衡的图像块;最后对各个集合中的图像块分别按相应准则进行分割.实验结果表明,对于均匀和非均匀光照条件下的文本图像,与全局分割算法、直接分块分割算法和Chou方法相比,该方法在分割效果上有显著提升,而且执行效率也较高.另外,对于部分非文本图像也同样有效.In order to overcome the two problems, parameters selection in the local thresholding segmentation methods which have many parameters usually and discontinuous problem among neighbor regions in the segmentation results, an interaetive document images thresholding segmentation algorithm based on image regions is proposed in this paper with the priori knowledge or experience from the users. Firstly, the presented method divides the image into several regions roughly. Secondly, it sorts all the image blocks according to their standard deviation values, which are taken as a measure to tell how much information of the background and object every block contains. Thirdly, the users input interactive information to separate all regions into three parts： blocks containing background or target only, blocks containing a small amount of background or target, and blocks containing distributing equilibrium background and target. Finally, the binarization of every block is conducted according to different criterion. Extensive experimental results show that the proposed scheme yields more promising binarization outcomes and also has better performance for document images under normal and inadequate illumination conditions, compared with global methods, simply thresholding approaches based on regions, and Chou＇s method. Moreover, the introduced approach is also effective for part of the non-text images.

关 键 词：图像分割 交互式文本图像分割 阈值选取 局部阈值 OTSU算法 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]