一种改进的基于最小生成树的遥感影像多尺度分割方法

影像分割是遥感影像面向对象信息提取的基础步骤。基于多特征、多尺度及考虑空间关系的遥感图像分割是主流研究方向。本文基于eCognition软件的多尺度分割思想,引入基于图论的最优化理论,提出了基于最小生成树分割和最小异质性准则的多尺度分割方法。该方法采用相干增强各向异性扩散滤波和最小生成树分割得到初始分割结果,通过最小异质性合并准则同时考虑多波段光谱特性区域形状参数进行区域合并,实现多尺度的影像分割。本次研究选取两景试验影像,对本文方法和eCognition软件的多尺度分割方法开展了目视比较和定量指标评价,结果表明,本文提出的方法是一种有效的影像分割方法,在光谱差异较小区域的细分方面优于eCognition方法。

An Improved Algorithm Based on Minimum Spanning Tree for Multi-scale Segmentation of Remote Sensing Imagery

As the basis of object‐oriented information extraction from remote sensing imagery ,image segmentation using multiple image features ,exploiting spatial context information ,and by a multi‐scale approach are currently the research focuses .Using an optimization approach of the graph theory , an improved multi‐scale image segmentation method is proposed .In this method ,the image is applied with a coherent enhancement anisotropic diffusion filter followed by a minimum spanning tree segmentation approach ,and the resulting segments are merged with reference to a minimum heterogeneity criterion T.he heterogeneity criterion is defined as a function of the spectral characteristics and shape parameters of segments .The purpose of the merging step is to realize the multi‐scale image segmentation .Tested on two images ,the proposed method was visually and quantitatively compared with the segmentation method employed in the eCognition software .The results show that the proposed method is effective and outper‐forms the latter on areas with subtle spectral differences .