泰森多边形区域高斯连接函数的遥感影像分割

Remote sensing image segmentation based on Voronoi regional Gaussian Copula

为了更好地描述多光谱遥感影像波段间的相关性,提高影像分割的精度,该文提出基于区域高斯连接函数的遥感影像分割方法。(1)利用泰森多边形划分技术将影像域划分成若干泰森多边形,在此基础上,利用马尔可夫随机场(MRF)模型对标号场进行定义;(2)引进高斯连接函数建立影像子区域像素光谱测度的多变量统计模型,以表达影像特征场光谱波段间的相关性;(3)定义各模型参数的先验概率,在贝叶斯定理的架构下实现影像分割模型的建立;(4)采用M-H算法对分割模型进行模拟,并利用最大后验概率(MAP)得到最优分割。利用所提算法分别对模拟影像和真实影像进行分割实验,结果表明:该文所提方法可以更准确地描述影像波段间的相关性,更有效地提高影像的分割精度。

In order to better describe the correlation between spectral remote sensing images and improve the precision of image segmentation,a regional Gaussian Copula based segmentation method for multispectral remote sensing image was proposed in this paper.(1)The image domain was divided into several Voronoi polygons by using Voronoi tessellation technology,Markov random field(MRF)model was used to define the labeling field based on the partition;(2)For sub-region images,Gaussian Copula was used to establish a multivariate statistical model of the image feature field to characterize the relativity among bands;(3)Based on the above labeling field,feature field model and the prior probability of parameters in the both model,a segmentation model was established under the framework of Bayes’theorem;(4)Metropolis-Hastings(M-H)algorithm was designed to simulate the segmentation model,and the optimal segmentation result could be obtained under the maximum a posterior(MAP)strategy.The segmented results showed that the proposed method could more accurately describe the correlation between image bands and improve the image segmentation accuracy.