摘要
为解决高分辨率遥感图像自动化处理程度不高的问题,提出一种基于邻接图的面向对象遥感图像分割方法.综合利用遥感图像的光谱信息和区域形状信息进行图像分割,并采用了一种新的异质性度量准则.与经典软件eCogniton在QuickBird图像分割的效率和效果方面的对比分析表明,该算法在运算效率上较eCognition的多尺度分割方法可以提高近1倍.
Automatically processing high-resolution remote sensing images is currently of regional and global research priority. This paper presented an algorithm based on adjacency graph partition for high-resolution remote sensing image segmentation. The proposed algorithm utilized both the region geometrical and spectral properties to evaluate the weight of the edges and the internal dissimilarity of the region. Comparing with the eCognition on image segmentation efficiency and effect, the proposed method can save half runtime in efficiency.
出处
《大连海事大学学报》
CAS
CSCD
北大核心
2009年第2期81-83,共3页
Journal of Dalian Maritime University
基金
空间数据挖掘与信息共享教育部重点实验室(福州大学)开放基金资助项目(200805)
极地测绘科学国家测绘局重点实验室开放基金资助项目(200810)
关键词
邻接图
面向对象
图像分割
高分辨率遥感
adjacency graph
object-oriented
image segmentation
high-resolution remote sensing
