中文摘要：

针对基于高斯混合模型的模糊聚类算法对噪声和异常值敏感的问题,利用包含邻域关系的先验概率与Student’s-T分布构建基于空间约束的混合模型.Student’s-T分布具有重尾的特点,较之高斯分布具有更强的抗噪能力.此外,在标号场上利用马尔科夫随机场模型刻画包含像素与其邻域像素相关性的先验概率,并表达为混合模型的权值系数以增强算法的鲁棒性.通过对合成图像和真实彩色图像分割结果的定性定量分析,验证了所提出算法的有效性和可行性.

英文摘要：

For the problem that the fuzzy clustering image segmentation algorithm based on Gaussian mixture model is sensitive to noises and outliers, a mixture model with spatial constraint is constructed by using a prior probability with neighborhood relationship and Student＇s-T distribution. The characteristic of heavy-tails in Student＇s-T distribution can overcome noise better than Gaussian distribution. In addition, the prior probability is constructed on the label field based on the interactions of pixel and its neighbors by markov random filed model, and is expressed as the weight degree in the mixture model to enhance robustness. The qualitative and quantitative analysis of the segmentation results for composite image and real color images show the effectiveness and feasibility of the proposed algorithm.