基于隐马尔可夫高斯随机场模型的模糊聚类高分辨率遥感影像分割算法

Hidden Markov Gaussian Random Field Based Fuzzy Clustering Algorithm for High-Resolution Remote Sensing Image Segmentation

本文利用隐马尔可夫随机场和高斯模型分别建立标号场和特征场的邻域关系,提出了基于隐马尔可夫高斯随机场模型的模糊聚类分割算法.该算法用隐马尔可夫随机场模型定义先验概率,并将该先验概率作为尺度控制因子引入到KL(Kullback-Lerbler)信息中,在目标函数的定义中,KL信息作为规则化项,其系数表示算法的模糊程度.在基于高斯模型的后验概率中,像素相关性被定义在空间和谱间,并用该概率的负对数值表征像素点到聚类中心的非相似性测度.通过对合成遥感影像和高分辨率遥感影像进行分割实验,证明了算法的有效性和普适性.

We establish neighbor relationships on both label field and feature field,and proposed an algorithm called the hidden M arkov Gaussian random field fuzzy c-means（ HM GRF-FCM） algorithm. In this algorithm,M arkov theory is used to define a prior distribution w hich is introduced to the KL（ Kullback-Lerbler） information and acts as a variable w hich controls the cluster size. KL information is used as a regularization item in the objective function and its coefficients can express the fuzziness of the algorithm. Besides,the posterior distribution is defined by Gaussian model w hich contains not only neighbor relationships in spatial space but also that in spectral space. Negative log posterior distribution function can express the dissimilarity betw een pixels and the cluster center,so it is defined as the dissimilarity measure. The algorithm is described in d-dimensional case,so w e do some segmentation experiments on 3-dimensional synthetic remote sensing image,3-dimensional high resolution IKONOS pan-sharpen images,and 4-dimensional SPOT-5 images to prove its accuracy and universality.