J4 ›› 2012, Vol. 29 ›› Issue (6): 657-664.

• 图像与信息处理 •    下一篇

基于Huber正则化二阶加速Richardson-Lucy湍流图像复原算法

邵慧1,2, 汪建业1, 徐鹏1, FDS团队   

  1. 1 中国科学院核能安全技术研究所;中国科学技术大学核科学技术学院;安徽合肥 230031;
     2 安徽建筑工业学院电子与信息工程学院,安徽 合肥 230601
  • 收稿日期:2012-02-20 修回日期:2012-05-09 出版日期:2012-11-28 发布日期:2014-03-20
  • 通讯作者: 汪建业(1966-)安徽人,研究员,从事光电子学测量技术和方法的研究 E-mail:jywang@aiofm.ac.cn
  • 作者简介:邵慧(1979-)女,安徽人,博士生,讲师,从事图像处理和系统控制的研究。E-mail:shaohui@aiai.edu.cn
  • 基金资助:

    国家863计划(2006AA861062);南京邮电大学“图像处理与图像通信”重点实验室开放课题(ZK209001)

Turbulence-degraded image restoration method using the second-order accelerated Richardson-Lucy algorithm based on Huber regularization

SHAO Hui1,2, WANG Jianye1, XU Peng1, FDS Team   

  1. 1 Institute of Nuclear Energy Safety Technology, School of Nuclear Science and Technology, University of Science and Technology of China,  Hefei  230031, China;
    2 School of Electronic and Information Engineering, Anhui University of Architecture, Hefei 230601, China 
  • Received:2012-02-20 Revised:2012-05-09 Published:2012-11-28 Online:2014-03-20

摘要:

为了快速准确地恢复湍流退化图像,提出了Huber正则化Richardson-Lucy(R-L)加速迭代盲反卷积(IBD)算法。根据图像滤波处理结果,采用Huber函数自适应地选择一阶范数和二阶范数正则化约束,增加算法收敛速度同时提高图像细节和边界复原质量。引入基于泰勒级数的二阶矢量外推加速方法,进一步增加迭代的收敛速度。实验结果表明,采用提议的算法需要的迭代次数较少,适用于实时性要求较高的场合,复原图像的主客观质量均有所提高。

关键词: 图像处理, 迭代盲反卷积, 矢量外推加速, Huber函数, 正则化技术

Abstract:

In order to restore turbulence-degraded images rapidly and exactly, an accelerated iterative blind deconvolution (IBD) Richardson-Lucy (R-L) algorithm based on Huber regularization is used. The Huber function can select L1 and L2 norms adaptively based on the processing result of image filter , in the smooth area ,the Huber function becomes the usual L2 least-squares penalty function with rapid convergence characteristics which can remove false edge ; and in the edge area, the L1 penalty function which can maintain the detail information and edge. Then the second-order vector extrapolation acceleration technique is used to accelerate convergence rate .The experimental results show that the proposed algorithm has greater convergence rate and better subjective and objective restoration result.

Key words: image processing, iterative blind deconvolution, vector extrapolation acceleration, Huber function, regularization technique

中图分类号: