Source-linked AI summary
Construction of Compactly Supported Shearlet Frames
P. Kittipoom, G. Kutyniok, W. Lim
TL;DR
Existing shearlet tight-frame studies focused on band-limited generators, despite the practical importance of spatial compact support. This paper derives frame conditions for irregular cone-adapted systems, constructs compactly supported frames using wavelet scaling functions and filters, and proves almost optimal sparse approximation of cartoon-like images.
Problem
Existing shearlet tight-frame studies focused on band-limited generators, while spatial compact support is important for practical applications.
Method
The paper derives sufficient frame conditions for irregular cone-adapted shearlet systems and constructs compactly supported frames using specifically designed wavelet scaling functions and filters.
Results
The constructed family has explicit frame-bound estimates and provides almost optimally sparse approximations of cartoon-like images.
Takeaways & Limitations
Cone-adapted shearlet frames can combine compact spatial support with frame stability estimates and almost optimal cartoon-like image approximation.
Takeaways & Limitations
Avoiding high redundancy requires balancing sampling constants, since choosing them too small causes significant computational complexity.
Abstract
from arXiv · showhide
Shearlet tight frames have been extensively studied during the last years due to their optimal approximation properties of cartoon-like images and their unified treatment of the continuum and digital setting. However, these studies only concerned shearlet tight frames generated by a band-limited shearlet, whereas for practical purposes compact support in spatial domain is crucial. In this paper, we focus on cone-adapted shearlet systems which -- accounting for stability questions -- are associated with a general irregular set of parameters. We first derive sufficient conditions for such cone-adapted irregular shearlet systems to form a frame and provide explicit estimates for their frame bounds. Secondly, exploring these results and using specifically designed wavelet scaling functions and filters, we construct a family of cone-adapted shearlet frames consisting of compactly supported shearlets. For this family, we derive estimates for the ratio of their frame bounds and prove that they provide optimally sparse approximations of cartoon-like images.
1. Introduction
The paper addresses the lack of practical cone-adapted shearlet frames with spatially compactly supported generators. It develops frame conditions and constructions that retain sparse approximation properties for cartoon-like images.
- Necessity of Compactly Supported Shearlets: Compact spatial support is important for precise geometric-feature detection and fast decomposition, but previous discrete shearlet approaches did not encompass this case.The paper also motivates compact support for adaptive schemes involving hyperbolic PDEs, stiffness matrices, boundary conditions, and fast algorithms.
- Contributions: The paper derives sufficient conditions for irregular cone-adapted shearlet systems to form frames, including explicit estimates for the ratio of their frame bounds.This extends earlier irregular-frame results beyond their dependence on band-limited generators and group-representation settings.
- Shearlet Background: Shearlets target anisotropic features through parabolic scaling and shear-based directional parameterization, supporting unified continuum and digital treatment.Their spatial footprint has size 2^-j times 2^-j/2, and shear rather than rotation supports the continuum–digital connection.
- Contributions: Using designed wavelet scaling functions and filters, the authors construct cone-adapted frames of compactly supported shearlets for horizontal-cone functions and then for L2(R2).The construction uses regular discrete systems and sufficiently small-determinant sampling matrices to obtain explicit frame-bound estimates.
- Sparse Approximation: The constructed compactly supported shearlet frames provide almost optimally sparse approximations of cartoon-like images.The remaining logarithmic factor is regarded as negligible compared with the N^-2 factor.
2. Regular and Irregular Shearlet Systems
Cone-adapted shearlets partition frequency space into directional cones and a low-frequency rectangle, supporting equal treatment of orientations. The paper formulates both regular and irregular discrete systems, with the latter allowing arbitrary discrete parameter sets.
- Cone adaptation: Discrete cone-adapted shearlets are the application-relevant variant because they treat continuum and digital settings uniformly.They are distinguished from systems arising directly from shearlet-group representations.
- Cone adaptation: Cone-adapted systems divide the frequency plane into four cones for directional components and a centered rectangle for low frequencies.The cone pairs C1/C3 and C2/C4 are treated separately.
- Regular systems: Regular systems sample scales, shears, and translations using a sampling vector c and indices j, k, and m.The shear range is |k| ≤ ⌈2^j/2⌉, with translations indexed by m ∈ Z2.
- Regular systems: Parabolic scaling produces anisotropic spatial footprints and corresponding anisotropic frequency regions at multiple orientations.The spatial dimensions are described as 2^-j × 2^-j/2, with frequency-domain regions shown by the tiling.
- Irregular systems: The paper generalizes regular sampling to irregular shearlet systems built from discrete sets Δ, Λ, and Λ̃ for scaling functions and two shearlet families.The associated transform records inner products with the scaling and shearlet elements.
- Generating shearlets: Feasible generators include classical shearlets and functions with sufficiently polynomially decaying Fourier transforms, including spatially compactly supported functions.Their frequency behavior is adapted to separable shearlet frames used later to construct compactly supported systems.
3. A General Sufficient Condition for Shearlet Frames
The paper develops sufficient conditions for cone-adapted irregular shearlet systems to form frames by controlling generator decay, overlap, and sampling geometry. It derives explicit frame-bound estimates and examines the trade-off between reconstruction quality and redundancy.
- General condition: The sufficient frame condition extends prior group-representation results to irregular parameters and potentially non-band-limited, compactly supported generators.The extension requires careful handling of the cone, scale, shear, and generator-decay relations.
- Analytic ingredients: The analysis uses a decay-adapted function together with overlap quantities that measure interactions among scaled and sheared generator supports.The quantities Lsup, R(c), and Γ support explicit estimates for frame bounds.
- Special estimates: The special shearing sequence sk = k yields a stronger estimate used later for compactly supported shearlet frames.This estimate is obtained after separately analyzing the k = 0 and k ≠ 0 cases.
- Frame theorem: Theorem 3.4 states that suitable bounds on these quantities ensure the irregular shearlet system is a frame for L2(C).The resulting frame bounds depend on the estimated upper and lower terms and the sampling geometry.
- Frame quality: The determinant of the sampling matrix represents the inverse translation-grid density and enters the frame-bound quotient B/A.A quotient closer to 1 corresponds to better frame reconstruction behavior.
- Frame quality: Smaller sampling constants can improve frame bounds but increase redundancy and computational complexity, so choosing sampling parameters is an application-dependent optimization.The paper identifies a trade-off between frame quality and the largest usable sampling constants.
4. Compactly Supported Shearlet Frames
The paper constructs compactly supported cone-adapted shearlet frames with explicit frame-bound estimates and proves their near-optimal sparse approximation of cartoon-like images.
- Construction: The construction begins with a maximally flat low-pass filter to build a separable, compactly supported shearlet generator with controllable frame bounds.The generator uses a wavelet-like horizontal function and a bump-like vertical function.
- Frame bounds: For sufficiently small sampling determinants, the constructed shearlet system has explicit positive lower and upper frame-bound estimates.The proof controls the lower bound through ˜Linf − R(c) > 0 and bounds the upper quantity Lsup.
- Frame bounds: There exists ˆc1 > 0 such that the system forms a frame for L2(C) when c2 ≤ c1 ≤ ˆc1.The admissible parameters also require c = (c1, c2) ∈ (R+)2.
- Frame-bound ratios: Theoretical frame-bound ratio estimates are reported for varied filter parameters and sampling constants, although numerical experiments perform better.The authors note that comparable estimates are unavailable even for compactly supported wavelet frames.
- Extension to L2(R2): The construction extends to L2(R2) using reflected shearlets and scaling functions, while preserving compact support in the spatial domain.The corresponding system forms a frame for L2(R2) under the same sampling constraint.
- Sparse approximation: The resulting compactly supported frames provide almost optimally sparse approximations for cartoon-like images.The approximation error is bounded by C · (log N)^3 · N^-2 as N → ∞.
5.1. Proofs of Results from Section 3.
The section proves explicit lower and upper frame-bound estimates by decomposing frequency-domain sums, applying Plancherel and inequalities, and estimating the resulting terms. These estimates establish frame existence for sufficiently small sampling-matrix determinants.
- Proof strategy: The proof decomposes the relevant sums over translations and frequency-scale cases before applying Plancherel’s theorem.The argument then resolves absolute values and uses the Cauchy–Schwarz inequality to obtain bounds.
- Frame bounds: An explicit estimate involving |det(Mc)| and ∥f̂∥2 yields the claimed frame-bound control.The proof concludes after deriving the final inequality and identifying the lower-bound contribution involving Linf − R(c).
- Proof strategy: The estimates split the analysis according to whether the maximum frequency component is |ω1| or |ω2| relative to |a_jξ1|.The second case is further divided into subcases for controlling the corresponding terms.
- Upper-bound estimate: The resulting bounds control the quantities used to estimate R(c), with separate estimates for the terms I1 and I2.The proof introduces constants T1, T2, and T3 and partitions index sets into Q and ˜Q when deriving the upper estimate.
5.2. Proofs of Results from Section 4.1.
The section analyzes trigonometric-polynomial filters through monotonicity, concavity, and derivative estimates, then uses these properties to obtain decay and feasibility conditions for the compactly supported shearlet construction.
- Filter analysis: The squared filter magnitude |m0|2 is represented using y = sin2(πξ1), and derivative analysis shows it decreases on the relevant interval.The proof also establishes concavity of |m0|2 on (0, 1/2).
- Filter analysis: For the modified filter ˜m0, the polynomial representation and derivative analysis establish monotonicity claims for cases K′ = 0 and K′ > 0.The argument studies the sign of (˜P)′ and the resulting behavior of ˜P and |˜m0(ξ1)|2.
- Proof completion: The remaining polynomial inequalities and coefficient estimates complete the proofs of the stated claims.These steps use bounds for ˜P and sums of coefficients before invoking the propositions that provide the final estimates.
- Feasibility: When K′ = 0 and L ≥ 10, the decay estimate gives γ > L/4 + 1/2 ≥ 3, implying the feasibility condition (6).This supplies the parameter condition required by the construction.