摘要
传统的基于动态规划的道路提取算法都是直接在图像域内根据道路的光谱等特征定义代价函数,当道路光谱特征发生变化时,需要重新定义新的代价函数,具有很大局限,不适用于道路特征复杂多样的高分辨率遥感影像。针对这一问题,提出了一种基于动态规划的道路中心线半自动提取算法:首先,利用阈值分割和核密度估计生成道路概率分布图;然后,根据道路概率分布图上的道路特征定义代价函数;最后,运用动态规划求解代价函数最大值来提取道路中心线。试验表明,提出的算法能够在高分辨率影像上提取各种不同光谱特征的道路中心线,取得了良好的效果。
The conventional road extraction algorithms based on dynamic programming all define merit function directly on the basis of road features on image domain. As the spectral characteristic of road varies,the conventional algorithms need to define merit function again, which is inflexible and unsuitable for high-resolution remote sensing image with various complex types of road. To address this problem, the road centerline extraction algorithm based on dynamic programming was proposed. Firstly, the thresholding and kernel density estimation are used to generate the road probability map,and then the merit function is defined based on road features in the road probability map.Lastly, the road centerlines are extracted by optimizing merit function using dynamic programming. Experimental results demonstrated that the proposed algorithm＇ s sability in extracting road centerlines of various spectral types from high-resolution remote sensing image.
出处
《测绘科学技术学报》
CSCD
北大核心
2015年第6期615-618,625,共5页
Journal of Geomatics Science and Technology
基金
国家自然科学基金项目(41001262)
关键词
高分辨率遥感影像
道路提取
道路中心线提取
动态规划
代价函数
high-resolution remote sensing image
road extraction
road centerline extraction
dynamic programming
merit function
