面向高分辨率卫星图像的几种图像分割算法稳定性对比分析  

Stableness Comparison of Image Segment Algorithm for High Resolution Remote Sensing Satellite Image

作　　者：康海洋[1,2] 何国金[1] 王杰生[1] 

机构地区：[1]中国科学院中国遥感卫星地面站重点实验室 [2]中国科学院研究生院,北京100039

出　　处：《科学技术与工程》2007年第16期4106-4110,4122,共6页Science Technology and Engineering

基　　金：国家自然科学基金项目(60272032);中国遥感卫星地面站创新课题(062103)资助

摘　　要：将基于均值漂移和图模型的图像分割算法及这两种算法的混合算法应用到高分辨率卫星光学图像,并对比了它们在分割遥感图像时的稳定性。均值漂移算法是一种基于核密度梯度估计的特征空间分析算法,其实质是一种统计优化过程。基于图模型的算法将一幅图像抽象为一个无向图,通过不断合并图结点,将这个图分割为多个连通分量,进而实现一幅图像分割。混合算法首先利用均值漂移技术对图像进行滤波,然后再使用图分割算法对图像进行分割。实验结果显示,均值漂移算法分割结果对其参数变化较为敏感,而基于图模型的算法和混合算法则较为稳定。Mean shift based image segmentation algorithm is applied. and the Hybrid of them to High Resolution Remote Sensing Satellite Image Graph Based image segmentation algorithm , also compares the stableness among these algorithms. Mean shift based image segmentation algorithm is a kind of kernel density estimation based feature space analysis algorithm, and the nature of it is statistical optimization. Graph based image segmentation algorithm treats the image as an undirected graph and combines graph＇s nodes to get connected components, at last gets segmented image. The hybrid algorithm first filters an image using mean shift technique and then segments the image with graph based segmentation algorithm. Experiments＇ results show that the mean shift based algorithm is more sensitive than the graph based algorithm and the hybrid algorithm, which are very stable to parameters and different images.

关 键 词：图像分割 均值漂移 图模型 分割稳定性 

分 类 号：TP751.1[自动化与计算机技术—检测技术与自动化装置]