基于信息聚类的遥感图像分割

发布时间：2018-02-09 06:55

本文关键词： 互信息聚类中心图像噪声相似性测度遥感图像分割　出处：《中国矿业大学学报》2017年01期 　论文类型：期刊论文

【摘要】：为解决经典聚类图像分割算法对聚类中心的依赖性和图像噪声的敏感性问题,提出了一种基于信息聚类的遥感图像分割方法.利用Gaussian分布建立同质区域内像素的概率分布,即假设每个同质区域内的像素都服从同一独立的Gaussian分布;结合Gaussian分布的特性建立像素对间灰度的联合分布.在此基础上,以互信息作为聚类算法的相似性测度,结合同质区域内以及同质区域间像素灰度的相似性建立目标函数,通过最大化求解上述目标函数,进而转化为迭代求解像素与同质区域的隶属度实现遥感图像分割.分别对模拟及真实遥感图像进行分割实验.结果表明:该方法不仅避免了聚类中心的选取,还降低了噪声敏感性,并且增强了图像分割的稳定性,从而验证了该方法的可行性及有效性.
[Abstract]:In order to solve the problem of the dependence of clustering center and the sensitivity of image noise in classical clustering image segmentation algorithm, a remote sensing image segmentation method based on information clustering is proposed. The probability distribution of pixels in homogeneous region is established by using Gaussian distribution. That is, assuming that the pixels in each homogeneous region are subject to the same independent Gaussian distribution, combining the characteristics of the Gaussian distribution, the joint distribution of gray levels between pixels is established. On this basis, mutual information is taken as the similarity measure of the clustering algorithm. Combining the similarity of pixel grayscale in homogeneous region and between homogeneous regions, the objective function is established, and the above objective function is solved by maximizing the objective function. Then it is transformed into iterative solution to solve the membership degree of pixel and homogenous region to realize remote sensing image segmentation. Experiments are carried out on simulated and real remote sensing images respectively. The results show that this method not only avoids the selection of clustering center, It also reduces the noise sensitivity and enhances the stability of image segmentation, which verifies the feasibility and effectiveness of the method.
【作者单位】：辽宁工程技术大学测绘与地理科学学院遥感科学与应用研究所;
【基金】：国家自然科学基金青年基金项目(41301479);国家自然科学基金面上项目(41271435) 辽宁省自然科学基金项目(2015020190)
【分类号】：TP751

【相似文献】