视觉灵敏性滤波和自适应正则化图像盲去模糊  

Blind image deblurring based on visual sensitivity filtering and adaptive regularization

作　　者：华夏[1,2,3] 张悦芮 刘广 费俊雄 HUA Xia;ZHANG Yuerui;LIU Guang;FEI Junxiong(Hubei Key Laboratory of Optical Information and Pattern Recognition,Wuhan Institute of Technology,Wuhan 430205,China;Hubei Research Center of Video Image and High Definition Projection,Wuhan Institute of Technology,Wuhan 430205,China;School of Electrical and Information Engineering,Wuhan Institute of Technology,Wuhan 430205,China)

机构地区：[1]武汉工程大学光学信息与模式识别湖北省重点实验室,湖北武汉430205 [2]武汉工程大学湖北省视频图像与高清投影研究中心,湖北武汉430205 [3]武汉工程大学电气信息学院,湖北武汉430205

出　　处：《华中科技大学学报（自然科学版）》2021年第11期64-70,共7页Journal of Huazhong University of Science and Technology(Natural Science Edition)

基　　金：国家自然科学基金资助项目(61801337,61671337)。

摘　　要：基于人类视觉系统(HVS)对模糊图像恢复的有效性,提出一种基于视觉灵敏度的图像盲去模糊先验(VSP).首先,通过统计实验和数学证明,模糊处理后的清晰图像的VSP值会减小,因此最大化VSP值有助于恢复更清晰的图像.然后,为了在模型迭代求解过程中保留图像的显著内容,对正则项施加自适应权重,在显著性强的地方使用小的惩罚系数,而在显著性弱的地方使用大的惩罚系数.最后,由于约束VSP的正则化项很难直接求解,因此采用基于半二次分裂的优化方案.研究结果表明:与现有的算法相比,本文模型在基准数据集和自然图像上都取得了较好的效果.Considering the effectiveness of the human visual system(HVS) in restoring blurred images,a blind image deblurring prior based on visual sensitivity(VSP) was proposed.First,through statistical experiments and mathematical proofs,the VSP value of sharp images after blur processing would be diminished,so maximizing the VSP value could help restore sharper images.Then,to preserve the salient content of the image in iterative solution process of the model,adaptive weight was applied to the regularization term.The small penalty coefficient was used where the saliency was strong,and the large penalty coefficient was used where the saliency was weak.Finally,as the regularization term of constrained VSP was difficult to be solved directly,an optimization scheme based on half-quadratic splitting method was adopted.Research results show that compared with the existing algorithms,the proposed model could achieve good results on both benchmark datasets and natural images.

关 键 词：人类视觉灵敏性 盲反卷积 模糊核估计 自适应正则化 显著性 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]