摘要
根据图像模糊产生方式的不同,研究人员将最为常见的模糊类型划分为离焦模糊和运动模糊两大类,而本文着重研究运动模糊的去除方法.在Krishnan和Fergus等人提出的基于超拉普拉斯先验的图像去模糊算法的基础上,本文给出了一种改进的基于超拉普拉斯约束的单幅图像去模糊算法.该算法主要分成三个处理步骤:在综合考虑图像边缘幅值和梯度的基础上,筛选出用于模糊核估算的图像子区域;对模糊核的稀疏性进行超拉普拉斯约束,快速且精确地估算出所需的模糊核信息;在对图像进行快速非盲反卷积复原阶段,原有的算法是用超拉普拉斯模型直接对图像梯度进行约束,而本文使用一种新的分布约束,从而生成了质量高,视觉效果好的去模糊图像.实验结果表明,本文算法可获得较好的图像去模糊效果,同时提高了去模糊算法效率.
According to the different ways of image blurring,researchers divide the most common fuzzy types into two categories： defocusing blur and motion blur,this paper focuses on the study of the method of removing motion blur. This paper presents an improved single image deblurring algorithm based on Hyper-Laplacian prior image deblurring algorithm proposed by Krishnan and Fergus et al.The algorithm is mainly divided into three steps：（ 1） Based on the Consideration of the amplitude and gradient of the image edge,the sub region of the image used to fuzzy kernel estimation is screened.（ 2） The improved Hyper-Laplacian constraint on the sparsity of the image fuzzy kernel is used to estimate the desired fuzzy kernel information of the image quickly and accurately.（ 3） In the fast non-blind deconvolution recovery stage,the original algorithm used the super-Laplacian model to constrain the image gradient directly,and this paper uses a new distribution constraint to produce deblurred image with high quality and nice visual effects. The experimental results show that the proposed algorithm can obtain better image deblurred effect and improve the efficiency of deblurring algorithm.
作者
秦绪佳
柯玲玲
范颖琳
郑红波
张美玉
QIN Xu-jia;KE Ling-ling;FAN Ying-lin;ZHENG Hong-bo;ZHANG Mei-yu(School of Computer Science and Technology, Zhejiang University of Technology, Hangzhou 310032, China)
出处
《小型微型计算机系统》
CSCD
北大核心
2018年第5期1097-1102,共6页
Journal of Chinese Computer Systems
基金
国家自然科学基金项目(61672462
61672463)资助
浙江省科技计划项目(2016C33165)资助
关键词
图像去模糊
超拉普拉斯
反卷积算法
运动模糊
image deblttrring
Hyper-Laplacian
deconvolution algorithm
motion blurring
