基于快速退火MRF的改进SAR图像分割方法

基于Markov随机场(MRF,Markov Random Field)的SAR图像分割方法利用了SAR图像的灰度和结构信息,能在分割过程中有效抑制斑点噪声,获得较高的分割精度.但这类方法的缺点是模拟退火的计算量很大.针对该问题,提出了一种基于快速退火MRF的SAR图像分割处理方法.该方法根据SAR图像Gibbs分布的特性,在求取全局最优解时,首先寻找邻域系统中占有支配地位的某种标记,若存在占支配地位的标记,用此标记更新状态;反之,则沿用传统模拟退火的方法随机更新状态.由于该方法引入基于Gibbs分布的先验判决进行系统状态更新,因此能够快速求得全局最优解.最后对真实SAR图像进行处理,处理结果验证了算法的有效性.

Modified SAR image segmentation method based MRF with fast simulated annealing

Markov random field (MRF) approaches are able to implement better SAR image segmentation by combing the image intensity and structure information. However, this kind of approaches often obtains a global solution at the cost of computational burden. A fast simulated annealing method was presented by combing the prior knowledge of SAR image distribution and applied to the SAR image segmentation based MRF. In the process of segmentation based MRF, the fast method first searched the neighborhood of each pixel to find out whether there was a predominant marker. If yes, the pixel would be marked with this predominant marker in the updated segmentation field; if not, the pixel would be marked randomly as the traditional simulated annealing algorithm does. Because a prior judgment based on Gibbs distribution was introduced, the proposed method is able to obtain a global best solution very quickly. In the end, experiments were carried out on real SAR images and the results validate the feasibility and efficiency of the proposed method.