基于区域特征的模糊多尺度Markov模型在纹理图像分割中的应用

Application of Texture Image Segmentation Based on the Fuzzy Multi-Resolution Markov Method with the Region-Based Feature

针对传统Markov模型中似然函数假设条件过于严格,观测图像像素间的相依关系不能充分利用的缺点,提出了一种基于区域特征的模糊多尺度Markov模型实现纹理图像分割模型.该模型首先利用一种区域特征提取方法,描述像素间的相依关系;然后,以区域特征的聚类结果作为先验信息,通过模糊多尺度Markov模型得到分割结果;最后采用Brodatz纹理库合成的人工图像作为实验数据,从定性和定量两方面验证了该模型的有效性.

In the traditional Markov random field model,the assumption of the likelihood function is very restrict and the relationship among the pixels of observed image can′t be effectively used.In order to overcome these disadvantages,this paper proposes a fuzzy multi-resolution Markov model based on the region feature for texture image segmentation.This model firstly describes the relationship among the pixels by using a scheme of region feature extracting.Then the segmentation is obtained by using the fuzzy multi-resolution Markov model,which uses the cluster result of region feature as the priori information.Finally qualitative and quantitative experiments demonstrates the validation of this model,where the test data are the synthetic images of the Brodatz texture database.