基于区域相似性与相异性的数字图像分割方法  

Digital image segmentation method based on regional similarity and diversity

作　　者：李楠[1] 徐书文[1] 

机构地区：[1]中国电子科技集团公司电子三所,北京100015

出　　处：《电视技术》2016年第7期24-27,55,共5页Video Engineering

摘　　要：针对数字图像数据量大、内容复杂、特征度量困难的特点,提出了一种综合区域相似性和相异性的基于图模型的分割方法。使用颜色方差作为距离度量,依靠区域邻接图和最近邻区域图来完成数字图像的快速区域合并分割。在合并过程中,通过合并区域的最小合并代价和最大合并代价变化,调整合并顺序,从策略上保证了分割区域的同质性和区域间的相异性。实验结果表明,该方法可以较好地解决图像的误分割现象。The data of digital image are huge quantity and very complex, so they are difficult to be described by the features. This paper proposes a digital image segmentation method based on the graph model which integrated regional similarity and diversity. The method uses color variance to define the distance between regions, and achieve the fast region merging based on region adja- cency graph and nearest neighbor graph. In the process of merging, the merge order adjusts with the change of the minimum and maximum cost of regional merge, so the homogeneity of regions and diversity between regions are ensured by the merge strategy. The experiments show that the proposed method is useful to solve the error segmentation.

关 键 词：图像分割 相似性 相异性 图模型 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]