[1]王杰,李胜光,宋一帆,等.图像去模糊的自适应交替方向乘子重叠组稀疏方法[J].郑州大学学报(工学版),2018,39(05):52-57.[doi:10.13705/j.issn.1671-6833.2018.05.017]
 Wang Jie,Li Shengguang,Song Yifan,et al.Image Deblurring using Adaptive Alternate Direction Multiplier Overlapping Group Sparsity Method[J].Journal of Zhengzhou University (Engineering Science),2018,39(05):52-57.[doi:10.13705/j.issn.1671-6833.2018.05.017]
点击复制

图像去模糊的自适应交替方向乘子重叠组稀疏方法()
分享到:

《郑州大学学报(工学版)》[ISSN:1671-6833/CN:41-1339/T]

卷:
39卷
期数:
2018年05期
页码:
52-57
栏目:
出版日期:
2018-08-21

文章信息/Info

Title:
Image Deblurring using Adaptive Alternate Direction Multiplier Overlapping Group Sparsity Method
作者:
王杰李胜光宋一帆白珂马天磊
郑州大学电气工程学院,河南郑州,450001
Author(s):
Wang Jie; Li Shengguang; Song Yifan; Bai Ke; Ma Tianlei
School of Electrical Engineering, Zhengzhou University, Zhengzhou, Henan 450001
关键词:
去模糊全变差重叠组稀疏ADMM自适应
Keywords:
Deblurring Total Variation Overlapping Group Sparse ADMM Adaptive
DOI:
10.13705/j.issn.1671-6833.2018.05.017
文献标志码:
A
摘要:
图像去模糊技术是图像处理领域的一个重要组成部分。由于重叠组稀疏全变差(OGSTV)正则化不仅具有保留边缘的特性,而且能够抑制阶梯效应的产生,正逐渐地应用到图像去模糊问题中。利用交替方向乘子(ADMM)方法来求解重叠组稀疏全变差模型时,其惩罚因子对去模糊问题的影响较大,且不易调节,故笔者在优化模型时根据复原出的图片自适应地调整该惩罚因子。该方法在保证计算速度的同时,自适应地复原出最佳图片,并保证了算法的鲁棒性。实验结果表明,本文在PSNR、SNR、相对误差等评价方法上均优于其它复原模型。
Abstract:
Image debluring technology played an important part in the image processing field. Total variables regularization with overlapping sparisity was gradually applied to the image deblurring problem. It could preserve image edge characteristics and suppress the generation of the staicase effect. When using the alternate direcction multiplier(ADMM) method to solve the overlapping group sparsity total variables model, the penalty factor could greatly influence the deblurring process and it was not easy to adjust. Therefore, a method was proposed to adaptively adjust the penalty factor according to the recovered image when the model was being optimized. This method adaptively restored the best picture and ensured the robustness of the algorithm while guaranteening the speed of calculation. Experimental results showed that the proposed method outperformed other recovery models in terms of PSNR, SNR, relative error and other evaluation indices.   
更新日期/Last Update: 2018-08-22