并行MCMC算法的SAR影像分割

MCMC(Markov Chain Monte Carlo, MCMC)方法采用顺序改变表征像素类属性的标号变量值会导致算法运算时间长、收敛速度慢等问题。为此，本文提出并行化改变像素标号值的MCMC方案，在贝叶斯推理框架下，依据高斯分布及MRF(Markov Random Field, MRF)模型建立SAR(Synthetic Aperture Radar, SAR)影像分割模型，设计实现基于多线程的并行采样方案；为了解决MRF标号场中邻域像素标号相关性问题，提出独立的像素并行采样的准则；同时，限制并行线程的数量，以保证采样的随机性。运用传统的串行算法和提出的并行算法对模拟和真实SAR影像进行影像分割实验；定性和定量的时间和精度评价结果表明:该方案在不影响分割精度的前提下大幅缩短影像分割时间，提高了效率。 

SAR Image Segmentation with Parallel MCMC Algorithm

The scheme of changing the pixels label variable values which are used to denoting the pixels class properties successively is usually used when using the MCMC (Markov Chain Monte Carlo, MCMC) algorithm, it usually leads to routine time consuming, slow speed of convergence. Therefore, a parallel MCMC scheme updating multiple pixels labels is proposed in this paper. In the framework of Bayesian inference, SAR (Synthetic Aperture Radar, SAR) image segmentation model is set up on the basis of Gaussian distribution and MRF (Markov Random Field, MRF) model. A parallel sampling scheme based on multithreading is designed; A independent pixel parallel sampling criterion is proposed so as to solve the label correlation problem between the neighborhood pixels labels in MRF label field; At the same time, the number of parallel threads is limited to ensure the sampling randomness. To verify the proposed parallel MCMC scheme, testing is carried out with real and simulated SAR images respectively by the serial algorithm and the proposed method. The results show that the proposed method behaves a significant reduction of running time without affecting the accuracy of segmentation.