基于类标号的快速RSST图像分割方法  

A RSST image segmentation method based on cluster label

作　　者：何兵兵[1] 程义民[1] 潘浩[1] 李杰[1] 

机构地区：[1]中国科学技术大学电子科学与技术系,安徽合肥230027

出　　处：《电路与系统学报》2009年第5期7-12,共6页Journal of Circuits and Systems

摘　　要：给出了一种基于类标号的快速回归最短生成树(RSST)图像分割方法。在传统RSST方法基础上,将区域相似性度量由像素域距离改进为类标号距离;通过定义类标号距离,可增加类标号图对应的加权图中权重为0的边所占比例,减少了传统方法中的区域合并次数,因此能较多的减少分割用时。方法已在微机上进行了模拟,结果表明,在相同分割数目和相当的分割效果情况下,分割用时明显减少。改进方法更适用于图像的实时分析和处理。The paper presents a recursive shortest spanning tree image segmentation method based on cluster label. Different from the traditional RSST method, the cluster label distance is used to measure the similarity of two regions instead of the pixel intensity distance in the new method, then the rate of zero weight edge rises in the graph mapped from the cluster label map, and the count of area combination decreases, finally the segmentation time reduces too. The result of experiments shows given the same area count the time for segmentation reduces greatly. The new approach can be used in real-time image analysis.

关 键 词：颜色量化 RSST方法 图像分割 类标号 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]