结合马氏距离的区域化模糊聚类遥感图像分割

发布时间：2018-02-02 11:53

本文关键词： 马氏距离 Voronoi多边形区域化模糊聚类遥感图像分割马氏距离规则化项　出处：《中国矿业大学学报》2017年01期 　论文类型：期刊论文

【摘要】：传统模糊聚类算法通常将像素光谱测度间的欧式距离作为相似性准则,其仅适用于同质区域内像素光谱测度呈对称分布的图像.同时,基于像素的图像分割算法极易受噪声干扰,导致该类算法难以适用于高分辨率遥感图像分割.因此,提出结合马氏距离的区域化模糊聚类遥感图像分割算法,该算法以Voronoi多边形为基本单元,令多边形内所有像素与该多边形具有同一类属性,并采用马氏距离定义非相似性测度.此外,在目标函数中引入马氏距离规则化项以控制聚类尺度,进而构建区域化高分辨率遥感图像分割模型.通过对合成及真实遥感图像分割结果的定性及定量分析,证明了提出算法具有较强的鲁棒性.
[Abstract]:The traditional fuzzy clustering algorithm usually regards the Euclidean distance between pixel spectral measures as similarity criterion, which is only suitable for symmetrical distribution of pixel spectral measures in homogeneous region. Pixel-based image segmentation algorithm is easily affected by noise, which makes it difficult to apply to high-resolution remote sensing image segmentation. Therefore, a regionalized fuzzy clustering remote sensing image segmentation algorithm combined with Markov distance is proposed. The algorithm takes the Voronoi polygon as the basic element, makes all pixels in the polygon have the same class of attributes as the polygon, and uses Markov distance to define the dissimilarity measure. The Markov distance regularization term is introduced into the objective function to control the clustering scale, and then the regionalized high-resolution remote sensing image segmentation model is constructed. The qualitative and quantitative analysis of the segmentation results of synthetic and real remote sensing images is carried out. It is proved that the proposed algorithm is robust.
【作者单位】：辽宁工程技术大学测绘与地理科学学院遥感科学与应用研究院;
【基金】：辽宁省自然科学基金项目(2015020090) 国家自然科学基金青年基金项目(41301479)
【分类号】：TP751
【正文快照】： 随着遥感传感器空间分辨率的提高,遥感图像中目标区域内像素光谱测度分布呈现分散、不规则以及非对称等特性.为了适应这些新特性,在高分遥感图像聚类分割算法设计中,不仅需要考虑像素的空间相关性,还需要建立更加有效的像素光谱测度相似性模型. 在众多图像分割算法中,模糊C

【相似文献】