摘要
针对遥感图像数据量大、地物对象与空间尺度密切相关的特点,提出一种自适应多尺度融合遥感图像分割方法。使用颜色方差作为距离度量,利用区域邻接图和最近邻区域图对遥感图像进行快速分割,建立阈值和尺度之间的函数关系,通过不同阈值得到多尺度分割结果,并采用融合方法获得最终结果。实验结果表明,与eCognition单尺度分割方法相比,该方法可消除遥感图像过分割或欠分割的现象。
For remote sensing image,the data are huge quantity,and the objects are closely related to spatial scale.This paper proposes an adaptive multi-scale integrated remote sensing image segmentation method.It uses color variance to define the distance between regions,and achieves the fast region merging based on region adjacency graph and nearest neighbor graph.Simultaneously,it establishes the relationship between scale and threshold,and then obtains multi-scale segmentation results based on different thresholds,and obtains the results by integrating multi-scale results.Experimental results show that the over-segmentation or sub-segmentation can be eliminated effectively.
出处
《计算机工程》
CAS
CSCD
2012年第24期208-210,215,共4页
Computer Engineering
基金
国家"973"计划基金资助项目(2012CB719903)
国家自然科学基金资助项目(41071256)
国家自然科学基金青年基金资助项目(41101386)
高等学校博士点基金资助项目(20090073110018)
关键词
遥感图像
图像分割
多尺度
图像融合
图模型
阈值
remote sensing image
image segmentation
multi-scale
image fusion
graph model
threshold