Abstract
We establish a construction of hybrid multiscale systems on a bounded domain \({\Omega } \subset \mathbb {R}^{2}\) consisting of shearlets and boundary-adapted wavelets, which satisfy several properties advantageous for applications to imaging science and the numerical analysis of partial differential equations. More precisely, we construct hybrid shearlet-wavelet systems that form frames for the Sobolev spaces \(H^{s}({\Omega }),~s\in \mathbb {N} \cup \{0\}\) with controllable frame bounds and admit optimally sparse approximations for functions which are smooth apart from a curve-like discontinuity. Per construction, these systems allow to incorporate boundary conditions.
Similar content being viewed by others
References
Bittner, K.: Biorthogonal spline wavelets on the interval. In: Wavelets and splines: Athens 2005, pp 93–104. Nashboro Press, Brentwood (2006)
Candès, E.J.: Ridgelets: Theory and applications. Stanford University, PhD thesis (1998)
Candès, E.J., Demanet, L.: The curvelet representation of wave propagators is optimally sparse. Comm. Pure Appl. Math. 58(11), 1472–1528 (2005)
Candès, E.J., Donoho, D.L.: New tight frames of curvelets and optimal representations of objects with piecewise C2 singularities. Comm. Pure Appl. Math. 57(2), 219–266 (2004)
Canuto, C., Tabacco, A., Urban, K.: The wavelet element method. I. Construction and analysis. Appl. Comput. Harmon. Anal. 6(1), 1–52 (1999)
Christensen, O.: An introduction to frames and Riesz bases. Applied and numerical harmonic analysis. Birkhäuser Boston, Inc, Boston (2003)
Cohen, A., Dahmen, W., DeVore, R. : Multiscale decompositions on bounded domains. Trans. Amer. Math Soc. 352(8), 3651–3685 (2000)
Cohen, A., Dahmen, W., DeVore, R.: Adaptive wavelet methods for elliptic operator equations: convergence rates. Math. Comp. 70(233), 27–75 (2001)
Cohen, A., Daubechies, I., Vial, P.: Wavelets on the interval and fast wavelet transforms. Appl. Comput. Harmon. Anal. 1(1), 54–81 (1993)
Dahlke, S., Fornasier, M., Raasch, T.: Adaptive frame methods for elliptic operator equations. Adv Comput. Math. 27(1), 27–63 (2007)
Dahlke, S., Raasch, T., Werner, M., Fornasier, M., Stevenson, R.: Adaptive frame methods for elliptic operator equations: the steepest descent approach. IMA J. Numer. Anal. 27(4), 717–740 (2007)
Dahmen, W.: Wavelet and multiscale methods for operator equations. Acta Numer. 6, 55–228 (1997)
Dahmen, W., Huang, C., Kutyniok, G., Lim, W.-Q, Schwab, C., Welper, G.: Efficient resolution of anisotropic structures. In: Extraction of quantifiable information from complex systems, pp. 25–51. Springer (2014)
Dahmen, W., Huang, C., Schwab, C., Welper, G.: Adaptive petrov–galerkin methods for first order transport equations. SIAM J. Numer Anal. 50(5), 2420–2445 (2012)
Dahmen, W., Kunoth, A., Urban, K.: A wavelet Galerkin method for the Stokes equations. Computing 56(3), 259–301 (1996)
Dahmen, W., Kutyniok, G., Lim, W., Schwab, C., Welper, G.: Adaptive anisotropic Petrov-Galerkin methods for first order transport equations. J. Comput. Appl. Math. 340, 191–220 (2018)
Dahmen, W., Schneider, R.: Wavelets with complementary boundary conditions—function spaces on the cube. Results Math. 34(3-4), 255–293 (1998)
Daubechies, I.: Ten lectures on wavelets, volume 61 of CBMS-NSF Regional Conference Series in Applied Mathematics. Society for Industrial and Applied Mathematics (SIAM), Philadelphia, PA (1992)
Dauge, M., Stevenson, R.: Sparse tensor product wavelet approximation of singular functions. SIAM J. Math Anal. 42(5), 2203–2228 (2010)
Demanet, L., Ying, L.: Wave atoms and time upscaling of wave equations. Numer. Math. 113(1), 1–71 (2009)
Dijkema, T.J., Stevenson, R.: A sparse Laplacian in tensor product wavelet coordinates. Numer. Math. 115(3), 433–449 (2010)
Donoho, D.: Sparse components of images and optimal atomic decompositions. Constr. Approx. 17(3), 353–382 (2001)
Donoho, D.L., Kutyniok, G.: Microlocal analysis of the geometric separation problem. Comm Pure Appl. Math. 66, 1–47 (2013)
Donovan, G.C., Geronimo, J.S., Hardin, D.P.: Intertwining multiresolution analyses and the construction of piecewise-polynomial wavelets. SIAM J. Math. Anal. 27(6), 1791–1815 (1996)
Donovan, G.C., Geronimo, J.S., Hardin, D.P.: Orthogonal polynomials and the construction of piecewise polynomial smooth wavelets. SIAM J. Math. Anal. 30(5), 1029–1056 (1999)
Easley, G., Labate, D., Negi, P.: 3D data denoising using combined sparse dictionaries. Math. Model. Nat Phenom. 8(1), 60–74 (2013)
Elad, M., Starck, J.-L., Querre, P., Donoho, D.L.: Simultaneous cartoon and texture image inpainting using morphological component analysis (MCA). Appl. Comput. Harmon. Anal. 19, 340–358 (2005)
Etter, S., Grohs, P., Obermeier, A.: FFRT - a fast finite ridgelet transform for radiative transport. Multiscale Model Simul. 13(1), 1–42 (2014)
Genzel, M., Kutyniok, G.: Asymptotic analysis of inpainting via universal shearlet systems. SIAM J. Imaging Sci. 7, 2301–2339 (2014)
Grohs, P., Obermeier, A.: Optimal adaptive ridgelet schemes for linear advection equations. Appl. Comput Harmon. Anal. 41(3), 768–814 (2016)
Guo, K., Kutyniok, G., Labate, D. : Sparse multidimensional representations using anisotropic dilation and shear operators. In: Wavelets and splines: Athens 2005, Mod. Methods Math, pp 189–201. Nashboro Press, Brentwood (2006) ,
Guo, K., Labate, D.: Optimally sparse multidimensional representation using shearlets. SIAM J. Math. Anal. 39(1), 298–318 (2007)
Harbrecht, H., Stevenson, R.: Wavelets with patchwise cancellation properties. Comp. Math. 75(256), 1871–1889 (2006)
Kittipoom, P., Kutyniok, G.: Construction of compactly supported shearlet frames. Constr. Approx. 35(1), 21–72 (2012)
Kutyniok, G., Lim, W.-Q.: Compactly supported shearlets are optimally sparse. J. Approx. Theory 163(11), 1564–1589 (2011)
Kutyniok, G., Lim, W.-Q., Reisenhofer, R.: ShearLab 3D faithful digital shearlet transforms based on compactly supported shearlets. ACM Trans. Math. Softw. 42(1), 1–42 (2016)
Labate, D., Lim, W.-Q., Kutyniok, G., Weiss, G.: Sparse multidimensional representation using shearlets. In: Proceedings SPIE (Wavelets and Sparsity XI), vol. 5914 (2005)
Nitsche, P.-A.: Best N term approximation spaces for tensor product wavelet bases. Constr. Approx. 24(1), 49–70 (2006)
Petersen, P.: Shearlet approximation of functions with discontinuous derivatives. J. Approx. Theory 207, 127–138 (2016)
Petersen, P.: Shearlets on Bounded Domains and Analysis of Singularities Using Compactly Supported Shearlets. Dissertation Technische Universiät Berlin (2016)
Petersen, P., Raslan, M.: Approximation properties of hybrid shearlet-wavelet frames for Sobolev spaces. Adv. Comp. Math., 2019, in press
Primbs, M.: Stabile biorthogonale Spline-Waveletbasen auf dem Intervall. Duisburg, Essen Univ., Dissertation (2006)
Stein, E.M.: Singular integrals and differentiability properties of functions. Princeton Mathematical Series, vol. 30. Princeton University Press, Princeton (1970)
Stevenson, R.: Adaptive solution of operator equations using wavelet frames. SIAM J. Numer. Anal 41(3), 1074–1100 (2003)
Voigtlaender, F.: Structured, compactly supported Banach frame decompositions of decomposition spaces. arXiv:1602.08772 (2016)
Acknowledgments
P. Petersen would like to thank Kristof Schröder and Massimo Fornasier for various discussions on related topics. Furthermore, P. Petersen expresses his gratitude to Massimo Fornasier and the Technische Universität München for the hospitality during P. Petersen’s research visit. Parts of this work was also done when J. Ma visited the Department of Applied Mathematics and Theoretical Physics of the University of Cambridge, and he is grateful for its hospitality. Parts of this work was accomplished while G. Kutyniok was visiting ETH Zürich. She is grateful to the Institute for Mathematical Research (FIM) and the Seminar for Applied Mathematics (SAM) for their hospitality and support during this visit.
Funding
J. Ma and P. Petersen acknowledge support from the DFG Collaborative Research Center TRR 109 “Discretization in Geometry and Dynamics”. G. Kutyniok acknowledges support by the Einstein Foundation Berlin, by the Einstein Center for Mathematics Berlin (ECMath), by Deutsche Forschungsgemeinschaft (DFG) SPP 1798, by the DFG Collaborative Research Center TRR 109 “Discretization in Geometry and Dynamics”, by the DFG Research Center Matheon “Mathematics for key technologies” in Berlin, and by the European Commission-Project DEDALE (contract no.665044) within the H2020 Framework Program.
Author information
Authors and Affiliations
Corresponding author
Additional information
Communicated by: Ivan Oseledets
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendices
Appendix A: \(H^{s}(\mathbb {R}^{2})\) Frame property of a reweighted shearlet system
This section shall be concerned with the proof of Theorem 3.3. The proof is split into Lemma A.2 for the upper frame bound and Lemma A.3 for the lower frame bound.
We use the following lemma to estimate weighted \(L^{2}(\mathbb {R}^{2})\) - frame coefficients by the \(H^{s}(\mathbb {R}^{2})\) norm of a function f.
Lemma A.1 (40, Lemma 2.4.9.)
Let\(s \in \mathbb {N}\), \(\tilde {C}>0,\)βsh > 3 and αsh > βsh + s. Further, let\(\phi , \psi , \widetilde {\psi } \in L^{2}(\mathbb {R}^{2})\)such that for almost all\(\xi \in \mathbb {R}^{2}\)there holds
Then there exists a constant C > 0 depending only on s, αshand βshsuch that\(\mathcal {SH}(\phi , \psi , \tilde {\psi }, c)\)satisfies
We continue to give the upper frame bound for the weighted shearlet system in \(H^{s}(\mathbb {R}^{2})\).
Lemma A.2
For\(s\in \mathbb {N}\)let\(\phi ^{1}, {\psi }^{1} \in L^{2}(\mathbb {R})\)be such that for some\(\tilde {C} \geq 0\), αsh − s > βsh > 3, and all 0 ≤ r1, r2 ≤ 2s,
We further denote ϕ = ϕ1 ⊗ ϕ1 and ψ = ψ1 ⊗ ϕ1and\(\tilde {\psi }(x_{1}, x_{2}) := \psi (x_{2}, x_{1})\)for all\(x\in \mathbb {R}^{2}\). Then, there exists some B > 0 depending only on αsh, βshand ssuch for\(({\psi }_{j,k,m,\iota })_{(j,k,m,\iota )\in {\Lambda }} = \mathcal {SH}(\phi , \psi , \tilde {\psi }, (c_{1},c_{2}))\)we have the estimate
for all\(f \in H^{s}(\mathbb {R}^{2})\).
Proof
We define for \(\mathbf {r}= (r_{1},r_{2})\): \(\phi ^{(\mathbf {r})} = D^{r_{1}}\phi ^{1} \otimes D^{r_{2}}\phi ^{1}\), \({\psi }^{(\mathbf {r})} = D^{r_{1}}{\psi }^{1}\otimes D^{r_{2}}\phi ^{1}\), and \(\tilde {\psi }^{(\mathbf {r})}(x_{1},x_{2})={\psi }^{(\mathbf {r})}(x_{2},x_{1})\). By Lemma A.1 there exists a constant C, dependent only on \({\alpha }_{\text {sh}}, {\beta }_{\text {sh}},\) such that for all \(0\leq r_{1},r_{2} \leq 2s\) and for \(({\psi }_{j,k,m,\iota }^{(\mathbf {r})})_{(j,k,m,\iota )\in {\Lambda }} = \mathcal {SH}(\phi ^{(\mathbf {r})}, {\psi }^{(\mathbf {r})}, \tilde {\psi }^{(\mathbf {r})}, c)\) we have that
From the assumptions (12) we have that \(\phi , \psi , \tilde {\psi } \in H^{2s}(\mathbb {R}^{2})\) and thus we can estimate the left-hand side of (13).
By using the product rule we calculate
with coefficients \(c_{\mathbf {r}}^{\mathbf {a}}\) bounded in absolute value by 22js. After applying (16) to (15) we estimate
Invoking (14), there exists some B > 0 such that
This yields the result. □
As a next step we provide the lower frame bound.
Lemma A.3
For\(s\in \mathbb {N}\), αsh − s > βsh > 4 and some\(\tilde {C}>0\)let\(\phi ^{1}, {\psi }^{1} \in L^{2}(\mathbb {R})\)be such that for all 0 ≤ r1, r2 ≤ 2s,
Further, let\(\phi , \psi , \widetilde {\psi }, \theta , \widetilde {\theta }, \mu \)be as in (1). Assume that there exists\(\bar c>0\)such that for all\(c_{1},c_{2}\leq \bar {c}\)we have that\(\mathcal {SH}(\mu ,\theta ,\tilde {\theta },(c_{1},c_{2}))\)forms a frame for\(L^{2}(\mathbb R^{2})\)with lower frame bound, which can be bounded from below independently ofc1, c2. Then there exists\(\tilde {c}>0\)such that for all\(c_{1}=c_{2} \leq \tilde {c}\)and c = (c1, c2) the system\(({\psi }_{j,k,m,\iota })_{(j,k,m,\iota )\in {\Lambda }} = \mathcal {SH}(\mu ,\theta ,\tilde {\theta },c)\), obeys
for all\(f\in H^{s}(\mathbb {R}^{2})\).
Proof
Let \(\bar {c} \geq c>0\). We define
By using the fact, that \(\mathcal {SH}(\mu ,\theta ,\tilde {\theta },c)\) is a frame for \(L^{2}(\mathbb {R}^{2}),\) we deduce
Since βsh > 4, there exists 𝜖 > 0 such that βsh − 1 − 𝜖 > 3 and for all \(\xi \in \mathbb {R}^{2}\)
Thus, we have that \((\theta _{j,k,m,\iota }- \theta _{j,k,m,\iota }^{c})_{(j,k,m,\iota ) \in {\Lambda }}\) satisfies the assumptions of Lemma A.2 with \(\sqrt {\tilde {C}} = 2c_{1}^{1+\epsilon }\). Hence, I2 can be estimated by \(c_{1}^{2+2\epsilon }/|\det {(M_{c})}| \|f\|_{H^{s}(\mathbb {R}^{2})}^{2}\) and since \(\det {(M_{c}}) = {c_{1}^{2}}\) this term is negligible for c1 small enough. Therefore we obtain
Now we substitute
Furthermore, we use that μc, 𝜃c, and \(\tilde {\theta }^{c}\) are all supported in \([-\frac {1}{2c_{1}},\frac {1}{2c_{1}}]^{2}\).
By using the Parseval identity we obtain
Now we substitute
to arrive at
Now set
Using the definition of \(\theta ^{c},\widetilde {\theta }^{c},\mu ^{c}\) then yields
We observe that
Thus we can estimate
Invoking Parseval’s identity again and reversing the transformations (19) as well as (18) from earlier shows that
By standard results on derivatives and the Fourier transform we have that
Thus we can invoke the Plancherel identity and partial integration to obtain
We still need to transition back from \({\psi }^{c}_{j,k,m,\iota }\) to ψj, k, m, ι. We proceed by invoking a triangle inequality
Using a similar estimate as for I2 in (17) we can estimate by invoking Lemma A.2 that IV2 is negligible for c small enough. This yields the result. □
To conclude this subsection we would like to examine how the auxiliary functions \(\theta ,\widetilde {\theta }\) and μ can be chosen such that they fulfill the frame property required in the proof of Theorem 3.3. Therefore assume \(\psi , \widetilde {\psi },\) and ϕ fulfill the assumptions of Theorem 3.3 with γ + 4 > αsh > γ > 4 + s. Then it follows that
Furthermore Ds(ψ1) and ϕ1 satisfy the assumptions of Theorem 3.3 and therefore there exists a sampling parameter \(\bar {c}>0\) such that for \(c_{1}=c_{2} \leq \bar c\) and c = (c1, c2) the system \(\mathcal {SH}(\phi , \psi , \widetilde {\psi }, c)\) constitutes a frame for \(L^{2}(\mathbb {R}^{2})\).
Appendix B: Localization of shearlet and wavelet frames
We now turn to the proof of Proposition 3.2. For this, we will require the following technical lemma.
Lemma B.1
[40] Let\(\psi \in L^{2}(\mathbb {R}^{2})\)be such that there exists C > 0 with
where β/2 > α > 1. Then, for ι = − 1, 1,
for a.e. \(({\xi }_{1}, {\xi }_{2}) \in \mathbb {R}^{2}\)and a constant\(C^{\prime }\).
Proof
See [40, Lemma 3.2.1] for a proof. □
Now we are ready to prove Proposition 3.2.
Proof
of Proposition 3.2 First of all, by the definition of \({\Theta }_{\tau , t}^{c}\) one has that for all \((j_{\text {sh}},k,m_{\text {sh}},\iota ) \in {{\Lambda }_{0}^{c}}\) and \((j_{\mathrm {w}},m_{\mathrm {w}},\upsilon ) \in {\Theta }_{\tau , t}^{c}\) such that \(j_{\mathrm {w}} < 1/(2\tau ) j_{\text {sh}} + t\) we have that \(supp (2^{-j_{\mathrm {w}}s}\omega _{j_{\mathrm {w}},m_{\mathrm {w}},\upsilon })^{d} \cap supp {\psi }_{j_{\text {sh}},k,m_{\text {sh}},\iota } = \emptyset \). Hence, we can assume in the sequel that \(j_{\mathrm {w}} > 1/(2\tau ) j_{\text {sh}} + t\).
Furthermore, the total number of wavelet translates for a fixed level jw is of order \(2^{2j_{\mathrm {w}}}\). W.l.o.g., the following computations can be done for υ = 1. Using this observation and Plancherel’s identity, we obtain
Leveraging on the frequency decay of the corresponding shearlet atoms, applying Lemma B.1, and using (W1) yields for any |a|≤ s
By a simple computation we obtain that if βw ≥ s
We plug this estimate into the estimate above and obtain with \({\beta }_{\mathrm {w}}^{\prime }={\beta }_{\mathrm {w}}-s\) and \({\beta }_{\text {sh}}^{\prime }={\beta }_{\text {sh}}/2-s\):
We continue by applying the substitution \(\xi \mapsto 2^{j_{\text {sh}}} \xi \)
Since \({\beta }_{\text {sh}}^{\prime }-{\alpha }_{\mathrm {w}}>1\) we obtain that the integral above is finite and hence we conclude
We rewrite the last sum above as
Since jw > 1/(2τ)jsh + t, we can now estimate
Since αw((1 − ε)/τ − 2) > 5/2 by assumption, the latter sum is finite. This leads to the estimate
Now (23) in combination with (22) and (21) implies together with (20) that
□
Rights and permissions
About this article
Cite this article
Grohs, P., Kutyniok, G., Ma, J. et al. Anisotropic multiscale systems on bounded domains. Adv Comput Math 46, 39 (2020). https://doi.org/10.1007/s10444-020-09784-0
Received:
Accepted:
Published:
DOI: https://doi.org/10.1007/s10444-020-09784-0