优化重加权L1范数的图像盲复原算法  

Blind Image Deblurring Based on Optimal Reweighted L1 Norm

作　　者：王凯[1] 肖亮[1,2] 黄丽丽[3] 韦志辉[1] 

机构地区：[1]南京理工大学计算机科学与工程学院,江苏南京210094 [2]江苏省光谱成像与智能感知重点实验室,江苏南京210094 [3]广西科技大学理学院,广西柳州545006

出　　处：《电子学报》2016年第9期2175-2180,共6页Acta Electronica Sinica

基　　金：国家自然科学基金(No.61171165;No.11431015;No.61302178);江苏省六大人才高峰项目(No.2012DZXX-036);中央高校基本科研业务费专项资金(No.30915012204);广西省自然科学基金(No.2014GXNSFAA118360)

摘　　要：在单幅运动模糊图像的盲复原问题中,图像中强边缘部分的利用成为模糊核估计的关键所在.为此,本文提出了一种优化重加权L1范数的图像盲复原算法.首先,建立了基于加权L1范数的模糊核盲估计模型,并引入了一种图像平滑模型对权重进行优化估计,从而减少计算权重时受细小结构以及噪声的影响,其次,设计了模糊核盲估计模型求解的迭代收缩阈值数值算法,最后采用了一种基于超拉普拉斯先验的快速图像非盲复原算法对模糊图像进行复原.仿真和实际数据实验结果验证了本文算法的有效性.In single blind motion deblurring,salient edges have been the key to success of kernel estimation. To this end,a newblind motion deblurring algorithm is proposed based on optimal reweighted L1 norm. Firstly,the weighted L1 based blind kernel estimation model is constructed. Then,for reducing the influence of noise and tiny structures,an image smoothing model is introduced into the optimal estimation of weights. A numerical algorithm based on iterative shrinkagethresholding is also proposed to solve the blind kernel estimation model. At last,a fast non-blind deconvolution method using Hyper-Laplacian priors is utilized to restore the final image. Experimental results on simulated and real-world data demonstrate the superiority of the proposed method.

关 键 词：图像盲复原 加权L1范数 迭代收缩阈值 

分 类 号：TP317.4[自动化与计算机技术—计算机软件与理论]