Robust uncertainty principles:: Exact signal reconstruction from highly incomplete frequency information

被引:11197
作者
Candès, EJ
Romberg, J
Tao, T
机构
[1] CALTECH, Dept Appl & Computat Math, Pasadena, CA 91125 USA
[2] Univ Calif Los Angeles, Dept Math, Los Angeles, CA 90095 USA
基金
美国国家科学基金会;
关键词
convex optimization; duality in optimization; free probability; image reconstruction; linear programming; random matrices; sparsity; total-variation minimization; trigonometric expansions; uncertainty principle;
D O I
10.1109/TIT.2005.862083
中图分类号
TP [自动化技术、计算机技术];
学科分类号
0812 ;
摘要
This paper considers the model problem of reconstructing an object from incomplete frequency samples. Consider a discrete-time signal f is an element of C-N and a randomly chosen set of frequencies Q. Is it possible to reconstruct f from the partial knowledge of its Fourier coefficients on the set Q? A typical result of this paper is as follows. Suppose that f is a superposition of vertical bar T vertical bar spikes f(t) = E-tau is an element of T f(tau)delta(t - tau) obeying vertical bar T vertical bar <= C-M (.) (logN)(-1 .) vertical bar ohm vertical bar for some constant C-M > 0. We do not know the locations of the spikes nor their amplitudes. Then with probability at least 1 - O(N-m), f can be reconstructed exactly as the solution to the l(1) minimization problem min/g Sigma(N-1)/t=0 vertical bar g(t)vertical bar, s.t. (g) over cap(omega) = (f) over cap(omega) for al omega is an element of ohm In short, exact recovery may be obtained by solving a convex optimization problem. We give numerical values for Cm which depend on the desired probability of success. Our result may be interpreted as a novel kind of nonlinear sampling theorem. In effect, it says that any signal made out of vertical bar T vertical bar spikes may be recovered by convex programming from almost every set of frequencies of size O(vertical bar T vertical bar (.) log N). Moreover, this is nearly optimal in the sense that any method succeeding with probability 1 - O(N-M) would in general require a number of frequency samples at least proportional to vertical bar T vertical bar (.) log N. The methodology extends to a variety of other situations and higher dimensions. For example, we show how one can reconstruct a piecewise constant (one- or two-dimensional) object from incomplete frequency samples-provided that the number of jumps (discontinuities) obeys the condition above-by minimizing other convex functionals such as the total variation of f.
引用
收藏
页码:489 / 509
页数:21
相关论文
共 32 条
[11]   A generalized uncertainty principle and sparse representation in pairs of bases [J].
Elad, M ;
Bruckstein, AM .
IEEE TRANSACTIONS ON INFORMATION THEORY, 2002, 48 (09) :2558-2567
[12]  
FENG P, 1996, P 1996 IEEE ISCAS, V2, P734
[13]  
Feng P., 1996, P IEEE INT C AC SPEE, V3, P1689
[14]   On sparse representation in pairs of bases [J].
Feuer, A ;
Nemirovski, A .
IEEE TRANSACTIONS ON INFORMATION THEORY, 2003, 49 (06) :1579-1581
[15]   On sparse representations in arbitrary redundant bases [J].
Fuchs, JJ .
IEEE TRANSACTIONS ON INFORMATION THEORY, 2004, 50 (06) :1341-1344
[16]  
Gilbert A C, 2002, P 34 ANN ACM S THEOR, P152
[17]  
GILBERT AC, 2004, UNPUB BEATING BOTTLE
[18]   Sparse representations in unions of bases [J].
Gribonval, R ;
Nielsen, M .
IEEE TRANSACTIONS ON INFORMATION THEORY, 2003, 49 (12) :3320-3325
[19]   MONOTONE CONVERGENCE OF BINOMIAL PROBABILITIES AND A GENERALIZATION OF RAMANUJANS EQUATION [J].
JOGDEO, K ;
SAMUELS, SM .
ANNALS OF MATHEMATICAL STATISTICS, 1968, 39 (04) :1191-&
[20]   EIGENVALUE DISTRIBUTION OF TIME AND FREQUENCY LIMITING [J].
LANDAU, HJ ;
WIDOM, H .
JOURNAL OF MATHEMATICAL ANALYSIS AND APPLICATIONS, 1980, 77 (02) :469-481