基于Gibbs随机场的有限混合模型改进与脑部MR图像的稳健分割

IMPROVEMENT OF FINITE MIXTURE MODEL BASED ON GIBBS RANDOM THEORY AND ROBUST SEGMENTATION FOR BRAIN MR IMAGES

有限混合 (FM )模型已经广泛地应用于图像分割 ,但是由于没有考虑空间信息 ,导致分割的结果对噪声很敏感 ,分割出的区域存在很多杂散的孤立点。本文在Gibbs随机场理论的指导下 ,将空间信息引入FM模型 ,提出了改进的脑部MR图像分割算法。此外 ,由树形K平均聚类来估计初始参数 ,实现了全自动的图像分割。本研究进行了仿真MR图像和真实MR图像的分割实验 ,定量的数据分析表明 ,我们所提的改进算法对噪声不敏感 ,可以更精确地将脑部MR图像标记为灰质。

The conventional finite mixture model (FM), being widely used in image segmentation, does not take the spatial information into account, which leads this model to work only on well-defined images. In this paper, in order to overcome this shortcoming of the FM model, we present an improved segmentation algorithm, based on Gibbs Random Field theory, for the segmentation of brain MR images. Furthermore, the new. algorithm is also an automatic one initialized by the tree-structure k-mean algorithm. The experiments on simulated MR images and real medical MR images prove that our new algorithm is insensitive to noise and can more precisely segment brain MR images into different tissues: gray matter, white matter and cerebrospinal fluid.