Majorization-minimization algorithms for wavelet-based image restoration

被引:422
作者
Figueiredo, Mario A. T. [1 ]
Bioucas-Dias, Jose M.
Nowak, Robert D.
机构
[1] Univ Tecn Lisboa, Inst Telecomunicacoes, P-1049001 Lisbon, Portugal
[2] Univ Tecn Lisboa, Inst Super Tecn, P-1049001 Lisbon, Portugal
[3] Univ Wisconsin, Dept Elect & Comp Engn, Madison, WI 53706 USA
关键词
image deconvolution; image restoration; majorization-minimization (MM) algorithms; optimization; regularization; wavelets;
D O I
10.1109/TIP.2007.909318
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Standard formulations of image/signal deconvolution under wavelet-based priors/regularizers lead to very high-dimensional optimization problems involving the following difficulties: the non-Gaussian (heavy-tailed) wavelet priors lead to objective functions which are nonquadratic, usually nondifferentiable, and sometimes even nonconvex; the presence of the convolution operator destroys the separability which underlies the simplicity of wavelet-based denoising. This paper presents a unified view of several recently proposed algorithms for handling this class of optimization problems, placing them in a common majorization-minimization (MM) framework. One of the classes of algorithms considered (when using quadratic bounds on non-differentiable log-priors) shares the infamous "singularity issue" (SI) of "iteratively reweighted least squares" (IRLS) algorithms: the possibility of having to handle infinite weights, which may cause both numerical and convergence issues. In this paper, we prove several new results which strongly support the claim that the SI does not compromise the usefulness of this class of algorithms. Exploiting the unified MM perspective, we introduce a new algorithm, resulting from using l(1) bounds for nonconvex regularizers; the experiments confirm the superior performance of this method, when compared to the one based on quadratic majorization. Finally, an experimental comparison of the several algorithms, reveals their relative merits for different standard types of scenarios.
引用
收藏
页码:2980 / 2991
页数:12
相关论文
共 53 条
[1]  
ANDREWS DF, 1974, J ROY STAT SOC B MET, V36, P99
[2]  
Andrews HC, 1977, DIGITAL IMAGE RESTOR
[3]   Regularization of wavelet approximations - Rejoinder [J].
Antoniadis, A ;
Fan, J .
JOURNAL OF THE AMERICAN STATISTICAL ASSOCIATION, 2001, 96 (455) :964-967
[4]  
Axelsson O., 1996, Iterative solution methods
[5]   Spatially adaptive wavelet-based multiscale image restoration [J].
Banham, MR ;
Katsaggelos, AK .
IEEE TRANSACTIONS ON IMAGE PROCESSING, 1996, 5 (04) :619-634
[6]   Wavelet domain image restoration with adaptive edge-preserving regularization [J].
Belge, M ;
Kilmer, ME ;
Miller, EL .
IEEE TRANSACTIONS ON IMAGE PROCESSING, 2000, 9 (04) :597-608
[7]   Bayesian wavelet-based image deconvolution: A GEM algorithm exploiting a class of heavy-tailed priors [J].
Bioucas-Dias, JM .
IEEE TRANSACTIONS ON IMAGE PROCESSING, 2006, 15 (04) :937-951
[8]   MONOTONICITY OF QUADRATIC-APPROXIMATION ALGORITHMS [J].
BOHNING, D ;
LINDSAY, BG .
ANNALS OF THE INSTITUTE OF STATISTICAL MATHEMATICS, 1988, 40 (04) :641-663
[9]  
BRIMBERG J, 2003, YUGOSL J OPER RES, V13, P199
[10]  
Burrus C.S., 1998, introduction to Wavelets and Wavelet Transforms-A Primer