基于广义模糊吉伯斯随机场图像分割新算法

New Approach for Image Segmentation Based on Generalized Fuzzy Gibbs Random Fields

吉伯斯分布作为一种引入图像空间信息的先验模型已广泛运用于贝叶斯图像分割中 .然而 ,由于传统该模型只在确定类上有定义 ,而在模糊类上未曾涉及 ,使得在运用该模型对一些模糊图像或退化图像进行处理时 ,分割效果不理想 ,甚至无能为力 .该文针对这些不足 ,从模型本身出发 ,在传统的吉伯斯随机场模型中引入模糊概念 ,并针对实际多值分割特点 ,提出一种高效、无监督的广义模糊算法 ,从而实现对多值图像的精确分割 .文中首先介绍一种二值的广义模糊吉伯斯随机场模型 ;然后将这种二值模型进行多值扩展 ,提出分段模糊与广义模糊吉伯斯两种实用的多值分割算法 ;最后将其运用于一系列医学图像分割 .实验表明 ,文中提出的广义模糊分割算法比基于传统随机场的算法有更好的图像分割能力 .

Gibbs distribution is a popular prior model widely used in Bayesian segmentation due to its excellent property describing the spatial information of image. However, the classical approaches without defining on fuzzy class may come across a lot of difficulties, such as getting the unexpected results or even nothing, when dealing with some fuzzy images or degraded one. In this paper, a fuzzy class is introduced into classical models of GRF to address these problems, and an efficient and unsupervised generalized fuzzy approach is presented to solve the case of multi-class, which makes the segmentation more precise. Firstly, the main contribution of this paper is to introduce a two-class model based on generalized fuzzy Gibbs random fields. Secondly, we extent the two-class fuzzy approach to a multi-class one, and two type of approaches are presented in one as Piecewise Fuzzy Gibbs Random Fields(PFGRF) and another as Generalized Fuzzy Gibbs Random Fields(GFGRF), which will be shown, in this paper, as a practical one. Finally, a series of images containing noise-free medical magnetic resonance image are presented in the experiments with the fuzzy approaches mentioned above. Our study shows that the approach based on GFGRF provide a powerful segmentation than other classical ones.