基于区域和统计的SAR分割方法,提出一种结合Voronoi划分技术、最大期望值EM (Expectation Maximization)和最大边缘概率MPM (Maximization of the Posterior Marginal)算法的多视SAR图像分割方法。首先利用Voronoi划分将图像域划分成不同的子区域,而每个子区域可以被看成待分割同质区域的一个组成部分,并假设每个子区域内的像素满足同一独立的Gamma分布,从而建立多视SAR图像模型,并在贝叶斯理论架构下建立图像分割模型,然后结合EM/MPM算法进行图像分割和模型参数估计。该方法将基于像元的马尔可夫随机场(Markov Random Field, MRF)模型扩展到基于区域的MRF模型,并且能同时有效地获取模型参数估计和基于区域的SAR图像最优分割。采用本文算法,分别对RADARSAT-Ⅰ/Ⅱ SAR强度图像和合成SAR强度图像进行了分割实验,定性和定量的测试结果验证了本文方法的有效性、可靠性和准确性。

We propose a novel multi-look synthetic aperture radar image segmentation method that combines Voronoi tessellation, expectation maximization (EM), and maximization of the posterior marginal (MPM) technology. The image domain is partitioned into a group of sub-regions by Voronoi tessellation, each of which is a component of homogeneous regions. Then a multi-look SAR image is modeled on the supposition that the intensities of pixels in each homogenous region satisfy an identical and independent gamma distribution. The image segmentation model is constructed based on the Bayesian paradigm. Finally, the EM/MPM algorithm, which integrates the EM algorithm for model parameter estimation and the MPM algorithm for image segmentation, is implemented. The proposed method expands pixel-based MRF to region-based MRF and achieves optimal segmentation and parameter estimation simultaneously. Results obtained from both real RADARSET-I/II and simulated SAR intensity images indicate that the proposed method is efficient and promising.