摘要
提出一种新的混合多阶段无监督图像分割算法。在第一阶段,通过分水岭算法得到一幅过度分割的图像,该图像中的所有小区域作为初始聚类状态将在接下来的层次聚类阶段中被合并。在第二阶段,一种新的启发式的基于Bayesian方法和Markov随机域的计算模型被用于基于区域的层次聚类算法,该算法用来合并初始分割结果中的邻接区域,以改进分水岭算法的分割效果。深入分析了该计算模型中两个相互作用的部分。通过对多种不同种类图像使用该算法进行分割,表明这种多阶段的方法适合无监督分割,它按照视觉一致的方式合并区域,并且比传统的层次聚类算法快很多。
A new combined multistage method for image segmentation was proposed. In the first stage, an oversegmented image was got using immersion watershed segmentation, and then primitive segmented results were provided for following merging. In the second stage, region based hierarchical clustering was used with a new heuristic computational model to merge adjacent primitive regions spatially. The model was derived from Bayesian method and Markov random field, and contained two interactive components. From experiments with three different kinds of images, the proposed method shows itself as a very effective way in unsupervised image segmentation. The hierarchy model merges regions in a way same as human perception and finishes within several seconds even for complex aerial image.
出处
《计算机应用》
CSCD
北大核心
2007年第3期673-676,共4页
journal of Computer Applications
基金
教育部优秀人才支持计划项目(NCET-04-0496)
教育部科学研究重点项目(105087)
关键词
分水岭算法
多阶段无监督分割
MRF
层次聚类
Bayesian方法
watershed
multistage unsupervised segmentation
Markov Random Field (MRF)
hierarchical clustering
Bayesian method