高如新, 朱新柳, 吴中华,等.联合正则化与低秩先验的自适应迭代盲图像复原[J].河南理工大学学报（自然科学版）,2023,42(5):137-143.

GAO R X, ZHU X L, WU Z H, et al.Adaptive iterative blind image restoration based on combining regularization and low rank prior[J].Journal of Henan Polytechnic University(Natural Science) ,2023,42(5):137-143.

联合正则化与低秩先验的自适应迭代盲图像复原

高如新1,2,3, 朱新柳1,2,3, 吴中华1,2,3, 谭兴国1,4

1.河南理工大学 电气工程与自动化学院，河南 焦作 454000;2.河南省煤矿装备智能检测与控制重点实验室，河南 焦作 454000;3.河南省智能装备直驱技术与控制国际联合实验室，河南 焦作 454000;4.哈密职业技术学院，新疆 哈密 839000

摘要:为改善运动模糊图像盲复原的效果，解决伪影显著、鲁棒性差、各尺度由于迭代次数固定而产生不利核估计的问题，提出一种联合正则化与低秩先验的自适应迭代盲图像复原算法。首先，利用l₀正则化先验的稀疏性估计中间复原图像和有效去除伪影，同时引入低秩先验抑制潜像恢复过程中的噪声干扰，提高模糊核估计的准确性；然后，针对多尺度迭代次数问题采用自适应策略，通过评估模糊核的相似性调整各尺度下的迭代次数；最后，用基于半二次分裂的交替优化策略求解本算法模型，利用非盲去模糊方法得到最终清晰图像。结果表明，本文算法能有效抑制噪声和伪影，鲁棒性好，并具有良好的复原效果。

关键词:图像盲复原;正则化;低秩先验;自适应迭代;模糊核估计

doi:10.16186/j.cnki.1673-9787.2021070053

基金项目:国家自然科学基金资助项目（61903126）；河南省科技攻关项目（212102210503）；新疆维吾尔自治区人才发展专项资金支持项目（202102）；新疆维吾尔自治区自然科学基金地州基金资助项目（2022D01F46）

收稿日期:2021/07/15

修回日期:2022/01/27

出版日期:2023/09/25

Adaptive iterative blind image restoration based on combining regularization and low rank prior

GAO Ruxin1,2,3, ZHU Xinliu1,2,3, WU Zhonghua1,2,3, TAN Xingguo1,4

1.School of Electrical Engineering and Automation，Henan Polytechnic University，Jiaozuo  454000，Henan，China;2.Henan Key Laboratory of Intelligent Detection and Control of Coal Mine Equipment，Jiaozuo  454000，Henan，China;3.Henan International Joint Laboratory of Direct Drive and Control of Intelligent Equipment，Jiaozuo  454000，Henan，China;4.Hami Polytechnic，Hami  839000，Xinjiang，China

Abstract:In order to improve the effect of blind restoration of motion blur images and solve the problems of obvious artifacts，poor robustness，and unfavorable kernel estimation due to fixed iteration number at various scales，an adaptive iterative blind image restoration algorithm based on combining regularization and low-rank prior was proposed.Firstly，the sparseness of l₀ regularized prior was used to estimate the intermediate restored image and effectively remove the artifacts.Meanwhile，a low-rank prior was introduced to suppress the noise interference in the process of latent image restoration and improve the accuracy of blur kernel estimation.Secondly，an adaptive strategy was adopted to evaluate the similarity of blur kernel to adjust the number of iterations at different scales.Finally，an alternative optimization strategy based on semi-quadratic splitting was used to solve the algorithm model，and the final clear image was obtained by non-blind deblurring method.Experimental results showed that the proposed algorithm can effectively suppress noise and artifacts，and had good robustness and good restoration effect.

Key words:blind image deblurring;regularization;low rank prior;adaptive iteration;blur kernel estimation

联合正则化与低秩先验的自适应迭代盲图像复原_高如新.pdf