Object-based Conditional Random Fields for Road Extraction from Remote Sensing Image

35th International Symposium on Remote Sensing of Environment

地球科学;生态环境科学

Huang, Zhijian^1,2 ; Xu, Fanjiang^2 ; Lu, Lei^3 ; Nie, Hongshan^1

SPDF, School of Electronic Science and Engineering, National University of Defense Technology, Changsha, China^1

IIST, Key Lab. Institute of Software, Chinese Academy of Science, Beijing, China^2

95980Troop, People's Liberation Army Air Force, China^3

关键词: Binary classification problems;    Conditional random field;    Contextual information;    Initial segmentation;    Object based image analysis (OBIA);    Remote sensing images;    Road extraction;    Topological information;   

Others  :  https://iopscience.iop.org/article/10.1088/1755-1315/17/1/012276/pdf
DOI  :  10.1088/1755-1315/17/1/012276

学科分类：环境科学（综合）

来源: IOP

【 摘 要 】

To make full use of spatially contextual information and topological information in the procedure of Object-based Image Analysis (OBIA), an object-based conditional random field is proposed and used for road extraction. Objects are produced with an initial segmentation, then their neighbours are constructed. Each object is represented by three kinds of features, including the colour, the gradient of histogram and the texture. Formulating the road extraction as a binary classification problem, a Conditional Random Fields model learns and is used for inference. The experimental results demonstrate that the proposed method is effective.