基于多核的MCMC图像去噪算法并行实现  

Parallel image denoising algorithm based on multi-core MCMC

作　　者：成敏[1,2] 张建州[1] 于岩[1] 

机构地区：[1]四川大学计算机学院 [2]中国人民解放军78098部队

出　　处：《计算机工程与应用》2014年第18期152-155,161,共5页Computer Engineering and Applications

摘　　要：图像去噪是许多图像处理任务的前提。马尔可夫链蒙特卡洛图像去噪算法是很重要的一种图像去噪方法,但去噪后图像存在明显斑点,在高噪声情况下去噪效果不理想,实际应用中需要进行噪声方差估计,运算速度慢。提出两步去噪方法,用均值滤波对噪声图像进行预处理,估计预处理后图像噪声方差,进行MCMC图像去噪;为充分利用多核处理器的硬件资源,研究了将MCMC算法进行并行编程,提高了程序的运行速度。实验表明两步去噪方法减少了斑点、提高了信噪比;并行实现提高了运算效率。Image denoising is a prerequisite for the many processing tasks of image. Markov Chain Monte Carlo algo-rithm is an important method of image denoising. However, the method has some problems such that the denoised image has obvious spots, the denoising image corrupted by heavy noise is not satisfactory, the noise variance needs to be estimated in practical application, and the operation speed of this method is slow. This paper proposes a two-step denoising method. It preprocesses the noise image using the mean filter. It estimates the pretreated image noise variance. It uses the MCMC image denoising method. To take full advantage of multi-core processor resources, this paper studies the parallel program-ming of MCMC algorithm. The multi-core program increases the speed of MCMC algorithm. The experiments show that the denoising method given in this paper reduces spots and improves the signal-to-noise ratio. Parallel processing can make the algorithm more efficient.

关 键 词：图像去噪 马尔可夫链蒙特卡洛方法(MCMC) 方差估计 预处理 并行处理 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]