一种新的基于变分模型的相干斑去噪算法研究  

Research on a New Variation Model Based Speckle Noise Despeckling Algorithm

作　　者：潘舟浩[1] 蔡爱华[1] 刘长清[1] 李鹏[1] 冯文森 

机构地区：[1]中国电子科学研究院,北京100041 [2]北京科技大学,北京100083

出　　处：《中国电子科学研究院学报》2016年第4期376-382,406,共8页Journal of China Academy of Electronics and Information Technology

摘　　要：专家场(Fields of Experts,FOE)图像先验模型是一种基于滤波器学习的高阶马尔可夫随机场(MRF)模型,对于许多图像复原问题该模型已经被验证其有效性。本文提出一种基于FOE图像先验模型的新的变分模型,用于相干斑噪声(乘性噪声)去噪。本文提出的变分模型需要求解一个非凸极小化问题,该问题可以通过i Piano(Inertial Proximal Algorithm for Nonconvex Optimization)算法来有效地解决。通过仿真图像和真实合成孔径雷达(Synthetic Aperture Radar,SAR)图像的去噪试验,可以表明本文提出的算法与目前最好的相干斑去噪算法性能相当。此外,本文提出的算法适用于图形处理器(Graphics Processing Unit,GPU)平台并行加速,可以大大提高运算效率。The fields of experts （FOE） image prior model is a filter-based high order Markov Random Fields model, the effectiveness of which is verified with regard to the image restoration problems. This paper proposes a FOE-based new variational model, which is used for speckle noise reduction. The pro- posed variational model needs to solve a non-convex minimization problem, which could be solved by iPi- ano algorithm. The results of experiment by simulated image and real SAR image, indicate that the per- formance of the proposed algorithm in this paper is comparable to that of the current most popular despeckling algorithm. Moreover, the proposed algorithm is suitable for parallel acceleration on GPU plat- form, which would significantly improve the efficiency of computation.

关 键 词：相干斑噪声 相干斑降噪 专家场 非凸优化 高阶马尔可夫随机场 

分 类 号：TN958[电子电信—信号与信息处理]