基于改进MRF的高分辨率SAR图像中建筑物轮廓提取算法

摘要

本文提出了一种基于改进马尔科夫随机场模型（MRF）的高分辨率SAR图像建筑物轮廓提取的方法。该方法首先引入了自适应的权重系数来改善邻域系统对先验能量项的影响，从而使分类结果更加准确。其次，利用Fisher分布来描述观测图像每一类的边缘分布，并且估计Fisher分布的参数；然后，根据改进的MRF对图像进行分类；最后，利用面向对象的方法，利用建筑物的形状特性及空间关系来提取建筑物。实验结果表明该方法可以较好地提取出建筑物的轮廓。%A novel method for the retrieve buildings from very high resolution synthetic aperture radar (SAR) imagery is pro-posed by the improved markov random field (MRF) model. The method we propose introduces the adaptive weight coefficient which improve the effect of neighborhood systems on the priori energy item, and makes the result of retrieve buildings more accurate. Firstly, a Fisher distribution for the marginal distribution of each class in the observed SAR image is employed and the parameters of a Fisher distribution are estimated using second-kind statistics (or Log-statistics). Then, images are classi-fied based on improved MRF. Finally, the buildings are extracted from SAR images by employing object-oriented methods, shape features of buildings and spatial relations. The experimental result demonstrates the effectiveness and efficiency of the proposed method.