摘要
在图像盲去模糊中,从单一模糊图像估计模糊核是个严重不适定问题.文中提出了一种基于稀疏先验和相对总变分的图像盲去模糊方法.该方法用权值L0平滑方法自适应地提取图像主体结构,剔除图像噪声、细节和小尺度物体边缘等不利于模糊核估计的因素;用相对总变分方法解决稀疏先验作为正则项估计复杂模糊核所存在的不准确性;用超拉普拉斯先验的正则化方法进行清晰图像估计.实验结果表明,文中算法相对于现有的图像去模糊方法,所估计出的清晰图像具有较好的结构和较少的伪迹,图像复原效果好.
In a blind image de-blurring,estimating a blur kernel from a single blurred image is a severely ill-posed problem. In this paper,a blind image de-blurring method based on sparse prior and relative total variation is proposed. In this method,a smoothing algorithm based on weighted L0 is employed to adaptively extract the main structure of an image and remove such factors adverse to a kernel estimation as the noise,the details and the edges of a small object,and a relative total variation method is adopted to overcome the inaccuracy of estimating a complex blur kernel by means of the regularization method of the sparse prior. Moreover,the regularization method of the Super-Laplacian prior is used to estimate the latent image. Experimental results show that,as compared with the existing image de-blurring method,the proposed method helps achieve the latent image of better structure and less artifacts,and it can better recover the latent image.
出处
《华南理工大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2017年第10期108-113,共6页
Journal of South China University of Technology(Natural Science Edition)
基金
广东省科技计划项目(2015B090909001)
广州市科技计划项目(2014Y2-00211)~~
关键词
权值L0
模糊核估计
相对总变分
运动去模糊
稀疏先验
显著边缘
weighted L0
kernel estimation
relative total variation
motion de-blurring
sparse prior
salient edge
