The Annals of Statistics

Recovering edges in ill-posed inverse problems: optimality of curvelet frames

Emmanuel J. Candès and David L. Donoho

Full-text: Open access


We consider a model problem of recovering a function $f(x_1,x_2)$ from noisy Radon data. The function $f$ to be recovered is assumed smooth apart from a discontinuity along a $C^2$ curve, that is, an edge. We use the continuum white-noise model, with noise level $\varepsilon$.

Traditional linear methods for solving such inverse problems behave poorly in the presence of edges. Qualitatively, the reconstructions are blurred near the edges; quantitatively, they give in our model mean squared errors (MSEs) that tend to zero with noise level $\varepsilon$ only as $O(\varepsilon^{1/2})$ as $\varepsilon\to 0$. A recent innovation--nonlinear shrinkage in the wavelet domain--visually improves edge sharpness and improves MSE convergence to $O(\varepsilon^{2/3})$. However, as we show here, this rate is not optimal.

In fact, essentially optimal performance is obtained by deploying the recently-introduced tight frames of curvelets in this setting. Curvelets are smooth, highly anisotropic elements ideally suited for detecting and synthesizing curved edges. To deploy them in the Radon setting, we construct a curvelet-based biorthogonal decomposition of the Radon operator and build "curvelet shrinkage" estimators based on thresholding of the noisy curvelet coefficients. In effect, the estimator detects edges at certain locations and orientations in the Radon domain and automatically synthesizes edges at corresponding locations and directions in the original domain.

We prove that the curvelet shrinkage can be tuned so that the estimator will attain, within logarithmic factors, the MSE $O(\varepsilon^{4/5})$ as noise level $\varepsilon\to 0$. This rate of convergence holds uniformly over a class of functions which are $C^2$ except for discontinuities along $C^2$ curves, and (except for log terms) is the minimax rate for that class. Our approach is an instance of a general strategy which should apply in other inverse problems; we sketch a deconvolution example.

Article information

Ann. Statist., Volume 30, Number 3 (2002), 784-842.

First available in Project Euclid: 6 August 2002

Permanent link to this document

Digital Object Identifier

Mathematical Reviews number (MathSciNet)

Zentralblatt MATH identifier

Primary: 62G20: Asymptotic properties 62C20: Minimax procedures
Secondary: 41A30: Approximation by other special function classes 42C40: Wavelets and other special systems 94A08: Image processing (compression, reconstruction, etc.) [See also 68U10] 94A12: Signal theory (characterization, reconstruction, filtering, etc.)

Ill-posed inverse problems regularization singular value decomposition Radon transform deconvolution edge edge-preserving regularization wavelets curvelets ridgelets wavelet-vaguelette decomposition wavelet shrinkage minimax estimation optimal rates of convergence white-noise model


Candès, Emmanuel J.; Donoho, David L. Recovering edges in ill-posed inverse problems: optimality of curvelet frames. Ann. Statist. 30 (2002), no. 3, 784--842. doi:10.1214/aos/1028674842.

Export citation


  • [1] BERTERO, M. (1989). Linear inverse and ill-posed problems. In Advances in Electronics and Electron physics (P. W. Hawkes, ed.). Academic Press, New York.
  • [2] BERTERO, M., DE MOL, C. and PIKE, E. R. (1985). Linear inverse problems with discrete data I: General formulation and singular sy stem analysis. Inverse Problems 1 301-330.
  • [3] BROWN, L. D. and LOW, M. G. (1996). Asy mptotic equivalence of nonparametric regression and white noise. Ann. Statist. 24 2384-2398.
  • [4] CANDÈS, E. J. (1998). Ridgelets: theory and applications. Ph.D. dissertation, Dept. Statistics, Stanford Univ.
  • [5] CANDÈS, E. J. (1999). Harmonic analysis of neural networks. Appl. Comput. Harmon. Anal. 6 197-218.
  • [6] CANDÈS, E. J. (1999). Monoscale ridgelets for image compression and denoising. Unpublished manuscript.
  • [7] CANDÈS, E. J. and DONOHO, D. L. (1999). Ridgelets: a key to high-dimensional intermittency? Philos. Trans. Roy. Soc. London Ser. A 357 2495-2509.
  • [8] CANDÈS, E. J. and DONOHO, D. L. (1999). Curvelets. Unpublished manuscript. Available at donoho/Reports/1999/curvelets.pdf
  • [9] CANDÈS, E. J. and DONOHO, D. L. (2000). Curvelets: A surpisingly effective nonadaptive representation of objects with edges. In Curve and Surface Fitting (A. Cohen, C. Rabut and L. L. Schumaker, eds.) 105-120. Vanderbilt Univ. Press, Nashville, TN.
  • [10] CHARBONNIER, P., BLANC-FÈRAUD, L., AUBERT, G. and BARLAUD, M. (1997). Deterministic edge-preserving regularization in computed imaging. IEEE Trans. Image Process. 6 298-311.
  • [11] DAUBECHIES, I. (1992). Ten Lectures on Wavelets. SIAM, Philadelphia.
  • [12] DAVISON, M. E. (1981). A singular value decomposition for the Radon transform in n-dimensional space. Numer. Funct. Anal. Optim. 3 321-340.
  • [13] DEANS, S. R. (1983). The Radon Transform and Some of Its Applications. Wiley, New York.
  • [14] DOBSON, D. and SANTOSA, F. (1996). Recovery of blocky images from noisy and blurred data. SIAM J. Appl. Math. 56 1181-1198.
  • [15] DONOHO, D. L. (1993). Unconditional bases are optimal bases for data compression and for statistical estimation. Appl. Comput. Harmon. Anal. 1 100-115.
  • [16] DONOHO, D. L. (1994). Statistical estimation and optimal recovery. Ann. Statist. 22 238-270.
  • [17] DONOHO, D. L. (1994). Asy mptotic minimax risk for sup norm loss: Solution via optimal recovery. Probab. Theory Related Fields 99 145-170.
  • [18] DONOHO, D. L. (1995). Nonlinear solution of linear inverse problems by wavelet-vaguelette decomposition. Appl. Comput. Harmon. Anal. 2 101-126.
  • [19] DONOHO, D. L. (1995). Denoising by soft-thresholding. IEEE Trans. Inform. Theory 41 613-627.
  • [20] DONOHO, D. L. (1997). Renormalizing experiments for nonlinear functionals. In Festschrift for Lucien Le Cam (D. L. Pollard, E. N. Torgersen and G. L. Yang, eds.) 167-181. Springer, New York.
  • [21] DONOHO, D. L. (1999). Wedgelets: nearly minimax estimation of edges. Ann. Statist. 27 859-897.
  • [22] DONOHO, D. L. (2000). Orthonormal ridgelets and linear singularities. SIAM J. Math. Anal. 31 1062-1099.
  • [23] DONOHO, D. L. (2001). Sparse components of images and optimal atomic decompositions. Constr. Approx. 17 353-382.
  • [24] DONOHO, D. L. and JOHNSTONE, I. M. (1994). Minimax risk over p balls for q losses. Probab. Theory Related Fields 99 277-303.
  • [25] DONOHO, D. L. and JOHNSTONE, I. M. (1994). Ideal spatial adaptation via wavelet shrinkage. Biometrika 81 425-455.
  • [26] DONOHO, D. L. and JOHNSTONE, I. M. (1995). Empirical atomic decomposition. Unpublished manuscript.
  • [27] DONOHO, D. L. and JOHNSTONE, I. M. (1998). Minimax estimation via wavelet shrinkage. Ann. Statist. 26 879-921.
  • [28] DONOHO, D. L. and JOHNSTONE, I. M. (1999). Asy mptotic minimaxity of wavelet estimators with sampled data. Statist. Sinica 9 1-32.
  • [29] DONOHO, D. L. and LIU, R. C. (1991). Geometrizing rates of convergence III. Ann. Statist. 19 668-701.
  • [30] DONOHO, D. L. and LOW, M. G. (1992). Renormalization exponents and optimal pointwise rates of convergence. Ann. Statist. 20 944-970.
  • [31] DONOHO, D. L. and NUSSBAUM, M. (1990). Minimax quadratic estimation of a quadratic functional. J. Complexity 6 290-323.
  • [32] EFROIMOVICH, S. YU. and PINSKER, M. S. (1981). Estimation of square-integrable density on the basis of a sequence of observations. Problemy Peredachi Informatsii 17 50-68 (in Russian). Problems Inform. Transmission 17 (1982) 182-196 (in English).
  • [33] EFROIMOVICH, S. YU. and PINSKER, M. S. (1982). Estimation of square-integrable probability density of a random variable. Problemy Peredachi Informatsii 18 19-38 (in Russian). Problems Inform. Transmission 18 (1983) 175-189 (in English).
  • [34] EFROMOVICH, S. and SAMAROV, A. (1996). Asy mptotic equivalence of nonparametric regression and white noise models has its limits. Statist. Probab. Lett. 28 143-145.
  • [35] FRAZIER, M. and JAWERTH, B. (1985). Decomposition of Besov spaces. Indiana Univ. Math. J. 34 777-799.
  • [36] FRAZIER, M. and JAWERTH, B. (1990). A discrete transform and decompositions of distribution spaces. J. Funct. Anal. 93 34-170.
  • [37] FRAZIER, M., JAWERTH, B. and WEISS, G. (1991). Littlewood-Paley Theory and the Study of Function Spaces. Amer. Math. Soc., Providence, RI.
  • [38] GEMAN, D. and REy NOLDS, G. (1992). Constrained restoration and the recovery of discontinuities. IEEE Trans. Pattern Anal. Mach. Intell. 14 367-382.
  • [39] GEMAN, D. and YANG, C. (1995). Nonlinear image recovery with half-quadratic regularization. IEEE Trans. Image Processing 4 932-946.
  • [40] HELGASON, S. (1980). The Radon Transform. Birkhäuser, Boston.
  • [41] IBRAGIMOV, I. A. and HAS'MINSKII, R. Z. (1984). Nonparametric estimation of the value of a linear functional in a Gaussian white noise. Theory Probab. Appl. 29 18-32.
  • [42] JOHNSTONE, I. M. and SILVERMAN, B. W. (1990). Speed of estimation in positron emission tomography and related inverse problems. Ann. Statist. 18 251-280.
  • [43] JOHNSTONE, I. M. and SILVERMAN, B. W. (1991). Discretization effects in statistical inverse problems. J. Complexity 7 1-34.
  • [44] JOHNSTONE, I. M. and SILVERMAN, B. W. (1997). Wavelet threshold estimators for data with correlated noise. J. Roy. Statist. Soc. Ser. B. 59 319-351.
  • [45] JONSSON, E., HUANG, S. C. and CHAN, T. (1998). Total variation regularization in positron emission tomography. Technical report, Dept. Mathematics, Univ. California, Los Angeles.
  • [46] KALIFA, J. and MALLAT, S. G. (1999). Thresholding estimators for inverse problems and deconvolutions. Unpublished manuscript.
  • [47] KHAS'MINSKII, R. Z. and LEBEDEV, V. S. (1990). On the properties of parametric estimators for areas of a discontinuous image. Problems Control Inform. Theory 19 375-385.
  • [48] KOROSTELEV, A. P. and TSy BAKOV, A. B. (1993). Minimax Theory of Image Reconstruction. Lecture Notes in Statist. 82. Springer, New York.
  • [49] LEMARIÉ, P. G. and MEy ER, Y. (1986). Ondelettes et bases hilbertiennes. Rev. Mat. Iberoamericana 2 1-18.
  • [50] LOUIS, A. K. (1986). Incomplete data problems in X-ray computerized tomography I. Singular value decomposition of the limited-angle transform. Numer. Math. 48 251-262.
  • [51] MALLAT, S. G. (1989). Multiresolution approximations and wavelet orthonormal bases of L2(R). Trans. Amer. Math. Soc. 315 69-87.
  • [52] MEy ER, Y. (1990). Ondelettes et Opérateurs I. Ondelettes. Hermann, Paris.
  • [53] MEy ER, Y. (1990). Ondelettes et Opérateurs II. Opérateurs de Calderón Zy gmund. Hermann, Paris.
  • [54] NUSSBAUM, M. (1985). Spline smoothing in regression models and asy mptotic efficiency in L2. Ann. Statist. 13 984-997.
  • [55] O'SULLIVAN, F. (1986). A statistical perspective on ill-posed inverse problems. Statist. Sci. 1 502-527.
  • [56] PILZ, J. (1986). Minimax linear regression estimation with sy mmetric parameter restrictions. J. Statist. Plann. Inference 13 297-318.
  • [57] PINSKER, M. S. (1980). Optimal filtering of square integrable signals in Gaussian white noise. Problemy Peredachi Informatsii 16 52-68 (in Russian). Problems Inform. Transmission 16 (1980) 120-133 (in English).
  • [58] RUDIN, L., OSHER, S. and FATEMI, E. (1992). Nonlinear total variation based noise removal algorithms. Phy s. D 60 259-268.
  • [59] STEIN, E. M. (1970). Singular Integrals and Differentiability Properties of Functions. Princeton Univ. Press.
  • [60] TEBOUL, S., BLANC-FÉRAUD, L., AUBERT, G. and BARLAUD, M. (1998). Variational approach for edge-preserving regularization using coupled PDEs. IEEE Trans. Image Processing 7 387-397.
  • [61] TERZOPOULOS, D. (1986). Regularization of inverse visual problems involving discontinuities. IEEE Trans. Pattern Anal. Machine Intell. 8 413-424.
  • [62] TIKHONOV, A. N. (1963). Solution of incorrectly formulated problems and the regularization method. Soviet Math. Doklady 4 1035-1039.
  • [63] VETTERLI, M. and KOVACEVIC, J. (1995). Wavelets and Subband Coding. Prentice-Hall, Englewood Cliffs, NJ.
  • [64] WAHBA, G. (1990). Spline Models for Observational Data. SIAM, Philadelphia.