Extraction of Road Network from High Resolution Remote Sensed Imagery with the Combination of Gaussian Markov Random Field Texture Model and  Support Vector Machine

WANG Min; LUO Jian-cheng; ZHOU Cheng-hu; MING Dong-ping; CHEN Qiu-xiao; SHEN Zhan-feng

doi:10.11834/jrs.20050340

Views : 0 下载量: 2547 CSCD: 25

PDF
Export
Share
Collection
Album

Extraction of Road Network from High Resolution Remote Sensed Imagery with the Combination of Gaussian Markov Random Field Texture Model and Support Vector Machine
Issue 3, Pages: 271-276(2005)
Published：2005
DOI： 10.11834/jrs.20050340
稿件说明：

移动端阅览

[1]汪闽,骆剑承,周成虎,明冬萍,陈秋晓,沈占峰.结合高斯马尔可夫随机场纹理模型与支撑向量机在高分辨率遥感图像上提取道路网[J].遥感学报,2005(03):271-276. DOI： 10.11834/jrs.20050340.

WANG Min, LUO Jian-cheng, ZHOU Cheng-hu, et al. Extraction of Road Network from High Resolution Remote Sensed Imagery with the Combination of Gaussian Markov Random Field Texture Model and Support Vector Machine[J]. Journal of Remote Sensing, 2005, (3): 271-276. DOI： 10.11834/jrs.20050340.

摘要

在高分辨率遥感图像上

道路网的同物异谱现象更为突出

因此其提取难度更大。提出了一种马尔可夫随机场纹理模型与支撑向量机分类相结合的道路网提取方法。其基本过程是:利用高斯马尔科夫随机场模型6个归一化特征值进行支撑向量机的分类得到道路斑块

利用形态学算子对其进行初步连接并提取轴线

然后通过斑块轴线的启发式连接得到最终道路网。试验证明方法是有效的。

Abstract

Extracting road network from high resolution remotely sensed imagery is much difficult than from the other resolution because high resolution imagery exhibits more complex spectral character.To distinguish them from other spa-tial objects

novel classification tools should be applied in which support vector machine（SVM）is outstanding in its fast training speed and strong capability in non-linear classification tasks.A road network extracting method combining Gaus-sian Markov random field texture model（GMRF）and SVM is proposed.This method can be divided into two main steps:firstly

GMRF is used to obtain the6texture features values of sample pixels

and SVM is trained and then used to clas-sify the whole image with these features into road patches vs.non road patches.After that

the patches are initially con-nected with some morphological operations

and their axes are extracted with thinning operation and then vectorized.Se-condly

a heuristic connecting strategy is used to connect and group the axes of the road patches into final road network.Experiments of road network extracting from IKONOS imagery validate our method.