基于区域MRF和贝叶斯置信传播的SAR图像分割

SAR Image Segmentation Using Markov Random Field Based on Regions and Bayes Belief Propagation

本文通过定义新的势函数,将贝叶斯置信传播算法和区域MRF模型有效结合,提出了一种SAR图像分割算法.考虑到SAR图像丰富的纹理信息,该算法对分水岭分割后的过分割区域提取纹理特征,在得到的区域邻接图上构建MRF模型,并加入区域灰度均值和方差作为区域特征,利用FCM聚类的初分割结果定义区域的关联势函数,并将区域特征引入到置信传播算法中,定义了新的交互势函数.该算法充分利用了SAR图像空间的背景信息,所定义的新的交互势函数能在促进分割结果区域一致性的同时较好保护边缘.实验结果表明,相对于其他MRF模型分割算法,本文算法能取得更好的分割效果.

Through defining the new potential functions,a SAR image segmentation method is proposed based on Bayes belief propagation and regional MRF model.Considering the rich texture information of SAR images,texture features are extracted from the watershed over-segmented regions,and then an MRF model is defined over the region adjacency graph of the initially segmented regions.Features of each small region are denoted by the texture features,the average and variance of the gray level of all the pixels in each region.The associated potential function is defined by the initial segmentation obtained from FCM clustering with the region.The features of the small regions are introduced to the interaction potential function.The new interaction potential function can effectively protect edge and promote regional consistency at the same time.In the experiments,the proposed algorithm is compared with other MRF image segmentation algorithms using real SAR images.The experimental results show that the proposed method is more effective for SAR image segmentation.