基于MRF模型的遥感图像建筑物分割研究

Object segmentation of remote sensing image based on MRF model

为了快速有效地从遥感图像中提取目标建筑物,采用小波分析和Markov随机场(MRF)相结合的方法进行遥感图像建筑物目标分割。首先将小波分解得到图像的多尺度序列作为各个尺度特征场的观测值,采用高斯混合模型建模特征场,以MRF模型作为标记场先验概率分布模型,通过EM算法迭代使得参数估计和图像分割交替进行,最后采用模板匹配检测建筑物目标的位置。选择多幅遥感图片进行实验,结果表明,采用分解层数为3的Haar小波,类别数为2,MLL模型势函数β为1时,该方法能够完成复杂背景下的建筑物目标分割并能对规则目标进行检测。

In order to extract the target buildings from rerate sensing images quickly and effectively, this paper presents an approach that combines wavelet analysis with Markov random fiel（MRF）. The multi-scale sequence by wawdet decomposition is regarded as the observation of feature field on each scale, which is modeled by gaussiau mixture model, and MRF as the prior probability distribution model of the label field. The parameter estimation and image segmentation split alternately via the EM algo- rithm iteration, and then template matching is used to detect the site of the object. Experinental analysis is carried out by selecting couples of remote sensing images, and the results show that when the level of Haar wavelet is 3, the number of categories is 2. the potential function of the MLL model β is 1, the method is able to complete the signentation of the goals of the buildings in tile conlplex eonteet and the detection of rule targets.