增广Lagrangian的快速视频复原方法  

Augmented Lagrangian based fast video restoration approach

作　　者：杨亚威[1,2] 李俊山[1] 芦鸿雁 张仲敏[1,4] 

机构地区：[1]第二炮兵工程大学信息工程系 [2]96215部队 [3]武警工程大学 [4]解放军西安通信学院

出　　处：《光电子．激光》2014年第11期2164-2169,共6页Journal of Optoelectronics·Laser

摘　　要：针对单幅模糊图像复原的局限性和视频应用的广泛性,提出了一种基于时空体和增广Lagrangian的快速视频复原方法。首先对视频复原与图像复原的特征进行比较和分析,研究视频复原的三维解卷积操作,并对目前存在的视频复原方法的实现过程与性能进行分析和总结;然后在时空体的思想下,通过时空联合的各向同性全变分来控制时间误差和空间误差,并引入一种增广的Lagrangian方法完成全变分规整化的难题;最后通过求解Lagrangian形式的f子问题和u子问题实现全变分最小化难题,并对规整化过程中的参数进行研究与讨论,最终实现了视频的快速鲁棒复原。基于仿真图像和实际视频的实验结果表明,本文方法的性能在运行时间和视觉质量评价方面都要优于当前的其它方法,能够有效地实现图像和视频的快速复原。Aiming at the limitation of single blur image restoration and the universality of video applica- tions, a fast and robust video restoration approach based on space-time volume and augmented Lagrang- Jan is proposed. Firstly, the characteristics of image restoration and video restoration are compared and analyzed,the three-dimensional deconvolution of video restoration operator is studied, and the realized process and performance of the existing video restoration methods are analyzed and summarized. And then,on the basis of the idea of space-time volume, the temporal error and spatial error is controlled through spatio-temporal isotropie total variation, and an augmented Lagrangian based method is intro- duced to implement the total variation regularization problem. Finally, the problem of total variation min- imization is achieved by solving the u sub-problem and f sub-problem in the form of Lagrangian,and the parameters are studied and discussed in the process of regularization, which achieves fast and robust vid- eo restoration at last. The experimental results based on both simulated images and real videos show that the performance of the proposed approach is superior to the current state of the art methods in the aspect of running time and several visual assessment indices,which can efficiently achieve the task of fast image and video restoration.

关 键 词：视频复原 时空体 增广Lagrangian 时空全变分 交替方向法 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]