借助EM 算法和模糊理论,提出了一种基于参数“软”估计和Markov随机场的SAR图像无监督分割方法。首先利用 多维空间的EM算法估计随机场的模型参数,并根据随机场模型参数分别计算观测数据的条件概率和标记图像的先验概 率,继而根据最大后验概率准则将图像分成具有相似统计特性的同质区域,重复以上步骤直至收敛。通过与传统的参数 “硬”估计分割算法的实验比较,该算法能更好保持图像边缘细节,区域连通性更好。

Using EM algorithm and fuzzy theory, an unsupervised SAR image segmentation method is proposed based on soft parameter-estimation and MRF model. First, parameters of the random fields are estimated by EM algorithm in multi-dimension fuzzy space, and the conditional probability of the observation data and the prior probability of the label image are calculated using the parameters estimated respectively, then MAP criteria is applied to partition the original image into several regions with similar statistical property. The procedure is repeated until it satisfy the convergent term. Compared with the traditional method based on the hard parameter-estimation, the detail of the original image is better preserved using the method proposed.