摘要
为了复原因相机抖动而产生的运动模糊图像,提出基于L p范数和全变分范数的正则化盲复原方法;首先,基于模糊图像的梯度稀疏性建立L p范数正则化模型,利用全变分范数保持图像的结构信息;然后,根据模糊核稀疏性的先验知识建立模糊核的盲估计模型;最后,提出一种渐近边界假设条件对模糊图像进行扩展以抑制振铃,并通过交替最小化方法分别求解清晰图像和模糊核的估计值。结果表明,所提出的方法简单、可行,具有更好的图像复原效果。
To restore the motion blurred images caused by camera shake,a regularized blind restoration method based on L p norm and total variation norm was proposed.Firstly,L p norm regularization model was established based on the gradient sparsity of blurred images,and the total variation norm was used to maintain the structure information of images.Secondly,a blind estimation model of the blur kernel was constructed according to the prior knowledge of the blur kernel sparsity.For suppressing the ringing effects,a asymptotic boundary hypothesis was finally proposed to extend the blurred images,and estimation of the latent image and the blur kernel were respectively solved by using the method of alternating minimization.The results show that the proposed method is simple and feasible,and has a better restoration effect of images.
作者
龚平
贺杰
刘娜
卢振坤
GONG Ping;HE Jie;LIU Na;LU Zhenkun(Guangxi Colleges and Universities Key Laboratory of Image Processing and Intelligent Information System,Wuzhou University,Wuzhou 543002,Guangxi,China;School of Electronics and Information Engineering,Wuzhou University,Wuzhou 543002,Guangxi,China;College of Computer Science and Electronic Engineering,Hunan University,Changsha 410082,Hunan,China;College of Computer Science,Chongqing Universtiy,Chongqing 400044,China;College of Electronic Information,Guangxi University for Nationalities,Nanning 530006,Guangxi,China)
出处
《济南大学学报(自然科学版)》
CAS
北大核心
2021年第1期16-23,共8页
Journal of University of Jinan(Science and Technology)
基金
国家自然科学基金项目(61562074,61961036)
广西创新驱动发展专项资金项目(科技重大专项)(桂科AA18118036)
广西自然科学基金项目(2018GXNSFBA281173,2018GXNSFAA294019)
广西高校中青年教师基础能力提升项目(2017KY0629,2018KY0537)。
关键词
盲复原
稀疏正则化
运动模糊图像
模糊核估计
边界条件
blind restoration
sparse regularization
motion blurred image
blur kernel estimation
boundary condition