武高玉++李慧云
DOI:10.16661/j.cnki.1672-3791.2016.27.137
摘 要:該文提出了一種新的應(yīng)用于圖像復(fù)原的加速動量梯度投影法。該方法在負梯度的方向上添加一個動量項,并且動態(tài)地選取動量參數(shù)和步長,從而加速了算法的收斂。在合理的假設(shè)下,證明了算法的全局收斂性。數(shù)值試驗表明,與當前先進的FISTA方法相比較,該文提出的算法無論是在時間上還是在圖像復(fù)原的質(zhì)量上都是有競爭力的。
關(guān)鍵詞:加速動量梯度投影法 動量 圖像復(fù)原
中圖分類號:TP391.41 文獻標識碼:A 文章編號:1674-098X(2015)09(c)-0137-04
A New Momentum Gradient Projection Method for Image Restoration
Wu Gaoyu1 Li Huiyun2
(1.School of Science Hebei University of Technology, Tianjin, 300401, China;2.School of Control Science and Engineering, Hebei University of Technology, Tianjin, 300401, China)
Abstract: In this paper, a new momentum gradient projection method for image restoration is proposed by using the convex combination of the negative gradient direction and the momentum term as the search direction, and the proposed method employs dynamic selection of momentum parameters and step length, which accelerates its convergence. Under mild conditions, the method is proved to be globally convergent. Experiment results demonstrate that the proposed method outperforms FISTA, both in time efficiency and in the quality of image restoration.
Key Words: Momentum gradient projection method; Momentum; Image restoration
表1是FISTA和算法1兩種算法圖像處理后的峰值信噪比(PSNR),運行時間(CPU) 的對比。從PSNR可以看出,用算法1復(fù)原的圖像與原始圖像最接近;從CPU可以看出,算法1速度較快。
圖1對2個測試圖像進行了圖像處理,將算法1與FISTA算法在圖像復(fù)原的質(zhì)量上進行了比較,可以看出,算法1復(fù)原的圖像的視覺效果稍微優(yōu)于FISTA算法復(fù)原圖像的視覺效果。
4 結(jié)語
該文提出了一種新的應(yīng)用于稀疏信號重構(gòu)的加速動量梯度投影法,即把負梯度方向與動量項的凸組合作為搜索方向。通過數(shù)值試驗的比較,該方法在圖像復(fù)原的質(zhì)量上與FISTA相當,但比FISTA收斂速度快,CPU時間更少。該文的方法是有效的。但其收斂速度還有待研究。
參考文獻
[1] Chambolle A,De Vore R A,Lee N Y,et al.Nonlinear wavelet image processing:variational problems, compression,and noise removal through wavelet shrinkage[J].Image Processing IEEE Transactions on,1998,7(3):319-335.
[2] Figueiredo M A T,Nowak R D.An EM algorithm for wavelet-based image restoration[J].IEEE Transactions on Image Processing A Publication of the IEEE Signal Processing Society,2003,12(8):16-906.
[3] Figueiredo M A T,Nowak R D.A bound optimization approach to wavelet-based image deconvolution[C]//IEEE International Conference on Image Processing. 2005:782-785.
[4] Beck A,Teboulle M.A fast iterative shrinkage-thresholding algorithm for linear inverse problems[J]. SIAM Journal on Imaging Sciences,2009,2(1):183-202.
[5] 劉紫娟,李慧云,劉新為. 外推系數(shù)帶參數(shù)的加速鄰近梯度算法[J]. 數(shù)值計算與計算機應(yīng)用,2016,37(3):211-222.
[6] Kim S J,Koh K,Lustig M,et al.An interior-point method for large-scale l1-regularized least squares[J].IEEE Journal of Selected Topics in Signal Processing,2007,1(4):606-617.
[7] Figueiredo M A T,Nowak R D,Wright S J.Gradient projection for sparse reconstruction: application to compressed sensing and other inverse problems[C]//IEEE Journal of Selected Topics in Signal Processing.2007:586-597.
[8] Ma G,Hu Y,Gao H.An accelerated momentum based gradient projection method for image deblurring[C]// IEEE International Conference on Signal Processing, Communications and Computing.2015.
[9] 畢亞倩,劉新為.求解界約束優(yōu)化的一種新的非單調(diào)譜投影梯度法[J].計算數(shù)學(xué),2013,35(4):419-430.