形状先验和图割的高分辨率遥感影像道路段提取  

Extracting Roads from High-resolution RS Images Based on Shape Priors and Graph Cuts

作　　者：周绍光[1] 陈超[2] 岳建平[1] 

机构地区：[1]河海大学地球科学与工程学院,江苏南京210098 [2]江苏省基础地理信息中心,江苏南京210013

出　　处：《测绘学报》2014年第1期60-65,共6页Acta Geodaetica et Cartographica Sinica

基　　金：国家自然科学基金(41271420);江苏省测绘科研项目(JSCHKY201201)

摘　　要：探索一种连接初始道路段的策略,并在该连接策略基础上研究从高分辨率影像中提取道路段的方法。道路提取流程包括:首先滤波预处理遥感影像获取待分割特征矢量;接着基于图割理论从待分割特征矢量中检测道路候选点;再通过直线匹配法处理道路候选点以获得初始道路段;最后基于图割理论和形状先验知识,并结合动态外推的思路连接初始道路段,以获得完整道路。试验证明了本文连接策略和道路提取算法的有效性。A road connection strategy and a road extraction method is studied from high resolution images. Here are the steps= firstly, filter images to form characteristics to be segmented; secondly, detect road candidate points from characteristics based on graph cuts theory; thirdly, process road candidate points with the line segment match method to achieve original road strips; finally, connect original road strips based on graph cuts theory, shape priors and dynamic segmentations to obtain whole roads. Experiments testify the effectiveness of the road connection strategy and the road extraction method respectively.

关 键 词：道路段提取 高斯混合模型 形状先验 图割 

分 类 号：P237[天文地球—摄影测量与遥感]