This paper presents a new algorithm for image segmentation, which combines Hidden Markov Random Field (HMRF) and Gaussian Regression Model (GRM) to Fuzzy C-Means (FCM) clustering. The proposed algorithm uses the KL (Kullback-Leibler) information to regularize the objective function of FCM, and then utilizes HMRF and GRM to model the neighborhood relationship of the label field and feature field, respectively. The HMRF model characterizes the neighborhood relationship through its prior probability, while the GRM is established under the assumption that a pixel has the same label with its neighbors. This paper takes some experiments with the proposed algorithm and other FCM based algorithms on the simulation image, real SAR image and texture image, respectively, and the accuracy of segmentation is evaluated. By comparing the results of them, the proposed algorithm can provided more accuracy segmentation result.%为了克服经典模糊聚类图像分割算法对图像噪声的敏感性，该文提出结合高斯回归模型(GRM)和隐马尔科夫随机场(HMRF)的模糊聚类图像分割算法。该算法用信息熵正则化模糊 C 均值(FCM)的目标函数，再用KL(Kullback-Leibler)信息加以改进，并将HMRF和GRM模型应用到该目标函数中，其中HMRF模型通过先验概率建立标号场邻域关系，而GRM模型则在中心像素标号与其邻域像素标号一致的基础上建立特征场邻域关系。利用提出的算法和其它经典算法分别对模拟图像、真实SAR图像以及纹理图像进行了分割实验，并对分割结果进行精度评价。实验结果表明，该文提出的算法具有更高的分割精度。
展开▼
