Abstract: The paper presents a rapid spatially constrained mixture model for image segmentation. For the spatial constraint, a filtering process is incorporated into the EM algorithm with only a small computational overhead of the standard EM algorithm. Images are down sampled to speed up segmentation process. In order to circumvent the problem of small sample model selection, the proposed spatially constrained model-based clustering method employs a weighted likelihood and assigns weight to each pixel according to the filtering process. The experiments show that the proposed image segmentation method can produce quick and reliable results comparable to the work using the original image size.

Key words: Image segmentation, EM algorithm, Weighted likelihood, Model selection

摘要： 采用滤波方法在EM算法中引入像素的位置信息，利用图像减采样方法以提高EM算法的收敛速度。为了避免小样本情况下混合分量选择的不稳定性问题，在所给出的受位置约束混合模型基础上，对采样数据进行加权处理。该方法在获得与原始分辨率分割效果相接近的情况下，能够明显地提高算法的运行速度。

关键词: 图像分割, EM 算法, 加权处理, 模型选择