HMGMRF模型及其在高分辨率影像分割中的应用

通过修改离差预测的方式，对高斯马尔可夫随机场(Gauss Markov Random Field)模型加以改进，提出层次型多光谱高斯马尔可夫随机场 (Hierarchical Multispectral Gauss Markov Random Field,HMGMRF) 模型及其相应的分割算法。影像分割时，先通过HMGMRF模型分析地物在各波段光谱特征的变化趋势（即地物各波段的纹理特征），期间结合了"谱间相关"这一特性，将离差预测时的邻域空间由原先的单层扩展为多层，增加了纹理特征的维度，从而提高了模型在描述纹理特征方面的能力；接着，基于贝叶斯原理，采用EM (Expectation Maximization)算法对各类地物的模型参数进行迭代估算；最后，基于增强型纹理特征，依据MAP (Maximum A Posteriori)原则，实现影像分割。实验结果表明，所提出的基于HMGMRF模型的分割算法具有较强的识别地物能力，可以获得较高的分割精度。

A HMGMRF Model and Its Application in High-resolution Imagery Segmentation

In this paper a Hierarchical Multispectral Gauss Markov Random Field (HMGMRF) model and its corresponding segmentation algorithm are proposed by modifying approach of anticipation dispersion of Gauss Markov Random Field (GMRF).In the segmentation procedure,the HMGMRF model is first used to analyze variational tendency of each land-cover classes in multispectral bands (i.e.multispectral texture characters of land-cover classes),neighborhood space is extended from single layer to multi-layer by introducing correlations of the spectral bands of remote sensing imagery,dimension of texture character is extended,thus capability to describe texture characters of the model is improved.Then,based on Bayesian principle,Expectation Maximization algorithm is accompanied by the estimation of model parameter on each land-cover classes.Finally,based on intensity texture characters,Maximum a posteriori is employed to perform image segmentation.Experimental results show that the proposed HMGMRF model-based segmentation algorithm is more capable in differentiating land cover classes and thus can achieve higher segmentation accuracy.