張凱 李敏
摘 要: 圖像去模糊一直是圖像修復(fù)中的重要問(wèn)題,針對(duì)經(jīng)典的去模糊方法,提出一種耦合非凸[lp(0≤p<1)]范數(shù)和G范數(shù)的圖像去模糊方法。該方法利用[lp(0≤p<1)]范數(shù)作為正則項(xiàng)約束,保證了圖像的稀疏性要求;利用G范數(shù)作為保真項(xiàng),保證在去模糊的同時(shí)有效抑制噪聲并保持圖像的細(xì)小特征,同時(shí)也給出新方法基于交替最小化的有效算法。實(shí)驗(yàn)結(jié)果表明新模型是可行的。
關(guān)鍵詞: 圖像去模糊; [lp(0≤p<1)]范數(shù); G范數(shù); 交替最小化
中圖分類號(hào): TN911.73?34 文獻(xiàn)標(biāo)識(shí)碼: A 文章編號(hào): 1004?373X(2016)05?0085?04
3 結(jié) 語(yǔ)
針對(duì)經(jīng)典的正則化去模糊方法,本文采用非凸[lp(0≤p<1)]范數(shù)作為正則項(xiàng)來(lái)保證圖像的稀疏性。同時(shí)選取G范數(shù)來(lái)刻畫噪聲成分,使得復(fù)原后的圖像含有較少的噪聲。對(duì)于耦合非凸[lp(0≤p<1)]范數(shù)和[G]范數(shù)的變分問(wèn)題,本文給出基于交替最小化迭代的算法。數(shù)值實(shí)驗(yàn)表明新算法是有效的。
參考文獻(xiàn)
[1] CHELLAPPA R, FAIN A. Markov random fields: theory and applications [M]. US: Academic Press, 1993.
[2] BIOUCAS?DIAS J M. Bayesian wavelet?based image deconvolution: a GEM algorithm exploiting a class of heavy?tailed priors [J]. IEEE transaction on image processing, 2006, 15(4): 937?951.
[3] RUDIN L I, OSHER S, FATEMI E. Nonlinear total variation based noise removal algorithms [J]. Physica D?nonlinear phenomen, 1992, 60(1): 259?268.
[4] BECK A, TEBOULLE M. A fast iterative shrinkage?threshol?ding algorithm for linear inverse problem [J]. SIAM journal on imaging sciences, 2009, 2(1): 183?202.
[5] OLIVEIRA J P, BIOUCAS?DIAS J M, FIGUEIREDO M A T. Adaptive total variation image deblurring: a majorization minimization approach [J]. Signal processing, 2009, 89(9): 1683?1693.
[6] MEYER Y. Oscillating pattern in image processing and nonlinear evolution equations [R]. Boston: American Mathematical Society, 2005.
[7] AUJOL J F, AUBERT G, BLANC?FERAUD L, et al. Image decomposition application to SAR image [C]// Proceedings of 2003 4th International Conference on Scale Space. Isle of Skye: Springer Berlin Heidelberg, 2003: 297?312.
[8] EKELAND I, TEMAM R. Analyse convexe and problems variationnels [M]. 2nd ed. French: Dunod, 1986.
[9] ZUO Wangmeng, MENG Deyu, ZHANG Lei, et al. A genera?lized iterated shrinkage algorithm for non?convex sparse coding [C]// Proceedings of 2013 International Conference on Computer Vision. Sydney: IEEE, 2013: 217?224.
[10] GILLES J, OSHER S. Bregman implementation of Meyer′s G?norm for cartoon+texture decomposition [R]. [S.l.]: UCLA CAM Report, 2001.
[11] KRISHNAN D, FERGUS R. Fast image deconvolution using hyper?Laplacian priors [C]// Proceedings of 2009 23rd Annual Conference on Neural Information Processing Systems. Vancouver: IEEE, 2009: 1033?1041.
[12] LAI M J, WANG J. An unconstrained [lp]minimization with 0<[p<1] for sparse solution of under?determined linear systems [J]. SIAM journal on optimization, 2009, 21(1): 82?101.
[13] KUANG?CHIH L, JEFFREY H, KRIEGMAN D J. Acquiring linear subspaces for face recognition under variable lighting [J]. IEEE transaction on pattern analysis machine intelligence, 2005, 27(5): 684?698.
[14] LEVIN A, FERGUS R, DURAND F, et al. Image and depth form a conventional camera with a code aperture [J]. ACM transaction on graphics, 2007, 26(3): 70?74.
[15] QIN L, LIN Z, SHE Y, et al. A comparison of typical [lp]mi?nimization algorithms [J]. Neurocomputing, 2013, 119(16): 413?424.
[16] MARJANOVIC G, SOLO V. On [lp]optimization and matrix completion [J]. IEEE transactions on signal processing, 2012, 60(11): 5714?5724.
[17] GOLDSTEIN T, OSHER S. The split bregman method for [L1] regularized problems [J]. SIAM journal on imaging sciences, 2009, 2(2): 323?343.
[18] OSHER S, YIN W, GOLDFARB D, et al. An iterative regularization method for total variation?based image restoration [J]. Multiscale modeling and simulation, 2005, 14(2): 460?489.