一种贝叶斯框架下的图像盲恢复算法  

Blind Image Restoration Algorithm in Bayesian Framework

作　　者：肖宿[1] 韩国强[1] 沃焱[1] 姚浩伟 

机构地区：[1]华南理工大学计算机科学与工程学院,广东广州510006 [2]深圳市华仁达电子有限公司,广东深圳518040

出　　处：《小型微型计算机系统》2010年第7期1443-1447,共5页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(60573019)资助;广东省重点实验室开放基金项目(CCNL-200704)资助;广东省科技计划项目(2008B080701052)资助;广东省科技计划项目(2009B030803004)资助;广东省自然科学基金博士科研启动基金项目(07300561)资助

摘　　要：基于各向异性扩散模型的图像盲恢复算法,其缺点在于速度难以得到显著地提升.为了加快图像恢复速度,同时尽可能提供令人满意的恢复结果.提出基于贝叶斯框架的图像盲恢复算法,首先引入调和模型作为原始图像和模糊的先验模型;然后,用伽马分布描述未知的参数;最后,利用变分近似的方法,以迭代的方式,交替地估计原始图像、模糊和参数的最优值.实验结果证明了该算法的有效性,与同类算法相比,可以得到更好的恢复结果;与基于各向异性扩散模型的算法相比,在速度上具有明显的优势.It is difficult to remarkably speed up the algorithms based on the anisotropic models,which is the inherent drawback in them.Therefore,this paper proposes a novel algorithm based on Bayesian framework to accelerate the image restoration with satisfactory restored results.Firstly,the harmonic model is introduce as the prior models of the original image and the blur;Secondly,the Gamma distribution is used to describe the unknown parameters;Finally,the variational approximation method is utilized to estimate the optimal original image,blur and parameters iteratively and alternately.The experimental results show the validity of the proposed algorithm which can obtain better restored results compared with the similar algorithms and has a remarkably advantage on the speed compared with the algorithms based on the anisotropic models.

关 键 词：图像恢复 贝叶斯框架 先验模型 调和模型 变分法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]