一种具有边缘保持的多尺度马尔可夫随机场模型图像分割方法  

Image Segmentation Method Using Multi-Resolution Markov Random Field Model with Edge-Preserving

作　　者：孟月波[1] 刘光辉[1] 徐胜军[1] 冯峰 MENG Yuebo;LIU Guanghui;XU Shengjun;FENG Feng(School of Information and Control Engineering, Xi’an University of Architecture and Technology, Xi’an 710055, China)

机构地区：[1]西安建筑科技大学信息与控制工程学院,西安710055

出　　处：《西安交通大学学报》2019年第3期56-65,共10页Journal of Xi'an Jiaotong University

基　　金：国家自然科学基金资助项目(51678470);国家重点研发计划合作单位项目(2017YFC0704207-03);陕西省自然科学基础研究基金面上项目(2015JM6276)

摘　　要：针对图像分割中常规四叉树结构的多尺度马尔可夫随机场模型非重叠区域在最优化过程中所造成的块效应,以及建模和推理过程导致低分辨率图像边缘细节模糊、缺失的现象,提出了一种具有边缘保持的多尺度马尔可夫随机场(Edge Preserving Multi-Resolution Markov Random Field, EPMRMRF)模型。该模型首先利用邻接区域之间的交互重叠约束,将局部区域的优化传递到相邻区域;其次采用具有边缘保持作用的Cauchy分布提取图像的多尺度边缘先验知识,在不同尺度上实现图像局部区域特征和多尺度边缘特征的融合。之后,为了对EPMRMRF模型进行迭代优化,提出一种分层区域置信度传播算法(Hierarchical Regional Belief Propagation Algorithm,HRBP),基于最大后验准则,求解马尔可夫随机场最大后验全局分布。实验结果表明,EPMRMRF模型和HRBP分割算法不仅有效保持了图像分割结果的边缘,获得了更好的分割结果,而且具有较快的分割速度,概率兰德指数相似性评价指标平均提升至0.890 9,全局一致性误差差异性评价指标平均降低至0.192 3。For conventional quad tree structure of multi-resolution Markov random field using in image segmentation, blocking effect is caused by optimization process in the overlap regions. Modeling and reasoning processes lead a low resolution image edge with fuzzy or lacking details. In this paper, an image segmentation model of edge-preserving multi-resolution Markov random field (EPMRMRF) is proposed. This model transfers the optimization of the local region into the neighboring region using interact overlap constraint between the adjacent regions. For the fuse of local regional character and multi-scale edge character in different scale images, the Cauchy distribution is utilized to extract image multi-scale edge prior knowledge. And then, a hierarchical regional belief propagation algorithm (HRBP) is proposed to iteratively optimize the EPMRMRF segmentation model by solving the Markov random field maximum a posteriori global distribution based on the maximum a posteriori. Compared with the conventional region-based Markov random field model, experiments show that the proposed algorithm can provide a better segmentation result and faster segmentation rate, in addition, the probabilistic rand index similarity evaluation index rises up to 0.890 9 and the global consistency error difference evaluation index drops to 0.192 3 on the average.

关 键 词：图像分割 多尺度马尔可夫随机场 边缘保持 分层区域置信度传播算法 

分 类 号：P391[天文地球—地球物理学]