基于字典对联合学习的退化图像复原方法  

A Degraded Image Restoration Approach Based on Pairs of Dictionaries Jointly Learning

作　　者：杨亚威[1,2] 胡双演[1] 张士杰 张姣 李俊山[1] 

机构地区：[1]第二炮兵工程大学信息工程系 [2]中国人民解放军96215部队

出　　处：《计算机辅助设计与图形学学报》2015年第3期406-413,共8页Journal of Computer-Aided Design & Computer Graphics

基　　金：国家自然科学基金(61175120)

摘　　要：针对传统字典学习方法在退化图像复原中效果不理想的问题,提出一种基于字典对联合学习的退化图像复原方法.首先在图像稀疏分解和字典学习的基本框架下,对基于字典学习复原方法的整个过程和关键步骤进行分析;然后针对图像复原的线性模型存在的缺陷,提出一种非线性的基于字典对联合学习的框架,解决了传统字典学习方法在退化图像复原中存在的不对称问题;最后利用随机梯度下降算法估计字典模型参数,并使用一种经典的启发式方法提高该算法的稳定性和收敛速度.基于各向同性和各向异性模糊核的实验结果表明,该方法对于非盲图像复原与当前技术条件下的方法相比是有竞争力的,甚至是更好的.A novel image restoration approach based on pairs of dictionaries jointly learning is proposed for the problem that the effect is weak to degraded images with traditional restoration approach based on dictionary learning. Firstly, the whole process and the key steps of restoration approach based on dictionary learning are analyzed in the basic frame of sparse decomposition and dictionary learning of images; And then, aiming at the limitation of the linear model of image restoration, a nonlinear frame based on pairs of dictionaries jointly learning is proposed, which solves the asymmetry problem of traditional dictionary learning technique in the process of degraded image restoration; Finally, the parameters of dictionaries model are estimated with a stochastic gradient descent algorithm, and the stability and speed of the algorithm is improved with a classical heuristic technique. The experimental results based on the isotropic and anisotropic kernels show that the proposed approach is competitive or even better than the state of the art approaches for non-blind image restoration.

关 键 词：图像复原 字典学习 稀疏分解 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]