An iterative thresholding algorithm for linear inverse problems with a sparsity constraint-英文文献.pdf

An iterative thresholding algorithm for linear inverse problems with a sparsity constraint Ingrid Daubechies Michel Defrise Christine De Mol February 11, 2008 3 0 0 2 Abstract l u We consider linear inverse problems where the solution is assumed to have a sparse ex- J pansion on an arbitrary pre–assigned orthonormal basis. We prove that replacing the usual 0 p quadratic regularizing penalties by weighted ℓ - penalties on the coefficients of such expansions, 1 p with 1 ≤ p ≤ 2, still regularizes the problem. If p 2, regularized solutions of such l -penalized ] problems will have sparser expansions, with respect to the basis under consideration. To com- A pute the corresponding regularized solutions we propose an iterative algorithm which amounts to F . a Landweber iteration with thresholding (or nonlinear shrinkage) applied at each iteration step. h We prove that this algorithm converges in norm. We also review some potential applications of t a this method. m [ 1 1 Introduction v 2 5 1.1 Linear inverse problems 1 7 In many practical problems in the sciences and applied sciences, the features of most interest cannot 0 3 be observed directly, but have to be inferred from other, observable quantities. In the simplest 0 approximation, which works surprisingly well in a wide range of cases and often suffices, there is a / h linear relationship between the features of interest and the observed quantities. If we model the object t a (the traditional shorthand for the features of interest) by a f