PaperlayerView exact revision on arXiv

Source-linked AI summary

Reconstruction and Range Characterization for a Directional Confocal Non-Line-of-Sight Imaging Model

Lingyun Qiu, Zuoqiang Shi, Jianyu Wang, Shiwei Wu

arXiv:2609.26846v1math.NAcs.NA

TL;DR

The paper asks what ideal confocal measurements determine about a volumetric directional field. It reduces the data to spherical means and reconstructs the irrotational component, proving that full data determine the divergence while divergence-free fields form the entire ambiguity.

  • Problem

    The paper asks what ideal confocal data determine about a volumetric directional field.

    Source: §1 · p. 2

  • Method

    Radial differentiation reduces the flux data to spherical means of the divergence, which a Fourier relation and Helmholtz multiplier use for explicit Fourier–sine reconstruction of the irrotational component.

    Source: §1 · p. 2

  • Results

    Full data determine the divergence uniquely, with divergence-free fields as the entire ambiguity; the irrotational representative is the unique minimum-Sobolev-norm field under the divergence constraint.

    Source: §1 · p. 2, §3.1 · p. 8

  • Takeaways & Limitations

    Measurements on an open relay-wall subset, for radii below a fixed bound, determine the divergence in the region covered by the corresponding wall-centered balls.

    Source: §1 · p. 3

    The analysis does not quantify finite-aperture, radial-truncation, or raw-noise effects; rendered and measured transients include unmodeled transport, so components are not surface-normal estimates.

    Source: §5.3 · p. 16, §6 · p. 18

Abstract

from arXiv · show

We study the reconstruction of a directional albedo field from confocal non-line-of-sight measurements. For mirror-symmetric vector fields in Sobolev spaces, radial preprocessing reduces the data to spherical means of the divergence with centers on the relay wall. Full data determine this divergence uniquely, with divergence-free fields forming the entire ambiguity. An explicit Fourier--sine formula recovers the irrotational Helmholtz component and reproduces the data. The Fourier--sine transforms of the model data form a weighted Hilbert space that we characterize exactly, with norm equal to the Sobolev norm of the reconstructed field. This weighted range remains well defined even when the preprocessed data fail to belong to standard Sobolev spaces. For Schwartz fields in the model class, such Sobolev regularity holds exactly when the divergence has zero depth integral. Measurements on an open subset of the relay wall, for all radii $0<r<R_{\max}$, uniquely determine the divergence in the union of the corresponding balls. An FFT-based algorithm with Stolt interpolation implements the reconstruction in $O(N^3\log N)$ operations on an $N^3$ grid. Tests on explicitly defined fields assess reconstruction accuracy and invariance under divergence-free perturbations. We present imaging examples of potential and vector-field reconstruction from model-generated, rendered, and measured transients.

1. Introduction.

The paper asks what ideal confocal measurements determine about a volumetric directional field. It reduces the data to a scalar divergence problem, characterizes reconstruction and its range, and establishes a partial-data uniqueness result.

1. Introduction.

  • Full confocal data determine the field's divergence, with divergence-free fields forming the entire ambiguity; a Fourier–sine formula recovers the irrotational component.That component is the least-Sobolev-norm field among those producing the same data.Source: §1 · p. 2
  • The transformed data form an exactly characterized weighted Hilbert range whose norm matches the reconstructed field's Sobolev norm, even when standard data regularity fails.For Schwartz fields, standard Sobolev regularity holds exactly when the divergence has zero depth integral.Source: §1 · p. 3
  • Measurements on an open relay-wall subset, for all radii below a fixed bound, determine the divergence in the region covered by the corresponding wall-centered balls.Source: §1 · p. 3
  • The implementation uses FFTs and Stolt interpolation, tests explicitly defined fields including divergence-free perturbations, and presents three examples from model-generated, rendered, and measured transients.Source: §1 · p. 3

2. Forward model.

The forward model extends the directional field across the relay wall with prescribed reflection symmetry, then converts transient measurements into weighted fluxes over wall-centered spheres. The divergence theorem reduces those data to a scalar spherical-mean problem, exposing the field's divergence-free ambiguity.

2.1. Function spaces and Fourier conventions.

  • The model extends the compactly supported directional-albedo field across the relay wall using reflection; the resulting vector field has even tangential components and an odd normal component.Its divergence is even in the depth coordinate.Source: §2.2 · p. 4, §2.1 · p. 4

2.2. From the transient to directional data.

  • With coincident virtual source and detector, transient data become negative outward fluxes through wall-centered spheres, weighted by r^-4.The conversion uses the time-to-radius relation and coarea formula.Source: §2.2 · pp. 4–5

2.3. Scalar reduction, gauge ambiguity.

  • The divergence theorem rewrites the measurement as a ball integral, so the data depend on the vector field only through its divergence.Consequently, adding any divergence-free field leaves the data unchanged.Source: §2.3 · p. 5
  • The full-data ambiguity consists of divergence-free fields, and oriented surface fields in the model are already irrotational.The stated mapping bound controls the data on any finite radius interval by the field's H1 norm.Source: §2.3 · pp. 5–6
  • Radial differentiation of the ball data yields spherical means of h = −∇·u, reducing reconstruction to a scalar problem.Source: §2.3 · p. 5

3. Full-data reconstruction and range characterization.

This section reconstructs the divergence-determined irrotational field from full data and characterizes the transformed data range through an isometric correspondence, distinguishing its reconstruction norm from standard Sobolev regularity governed by depth cancellation and cone-boundary behavior. It also establishes that both standard-data regularity regimes occur in the model class, that depth cancellation gives no uniform reverse norm estimate, and that partial measurements yield local uniqueness for the divergence.

3.1. Canonical reconstruction.

  • The Fourier relation reconstructs h = −∇·u from the data, and the resulting irrotational field equals the Helmholtz projection and reproduces those data.Its Fourier transform is given by 2|k3|Θγ(k′, |k|).Source: §3.1 · pp. 7–8
  • Two fields produce the same full data exactly when their difference is divergence-free; the canonical reconstruction is the unique minimum-norm representative.It minimizes the Hs norm both among fields with the reconstructed divergence and among fields producing the data.Source: §3.1 · p. 8

3.2. Isometry and the exact transformed range.

  • The Fourier–sine transform maps the data space isometrically onto its weighted Hilbert range, with inverse reconstruction from that transformed range.The range space is complete, so the data space is also a real Hilbert space.Source: §3.2 · pp. 10–11

3.3. Standard Sobolev data norms and depth cancellation.

  • For Schwartz fields, the preprocessed data belong to Hs exactly when the divergence has zero depth integral; otherwise they fail to belong for every s.The standard Sobolev data weight also has an unbounded ρ/q factor near the cone boundary, while excluding that boundary restores norm comparison.Source: §3.3 · pp. 12–13
  • The stated norm ratio follows from a weighted-average argument over supp F_n, giving H^s N_{δ,s}(γ_n)^2 = δ/(2 − δ^2)^s.The pointwise ratio converges uniformly on supp F_n to δ/(2 − δ^2)^s.Source: §3.3 · p. 13
  • Mollified surface fields have nonzero h̄ = −Δ′ϕ̄ε yet transformed data in every Y^s, and they are reconstructed exactly.Their compactly supported ϕ̄ε is nonzero and therefore cannot be harmonic.Source: §3.3 · pp. 13–14
  • Even within the depth-cancelling Schwartz class, the standard Sobolev data norm has no uniform reverse estimate by the reconstruction norm.Depth cancellation characterizes finiteness of the standard data norm, but not its uniform equivalence with the reconstruction norm.Source: §3.3 · p. 14
  • The scalar reduction yields local uniqueness for the divergence from measurements on an open region Γ, for a fixed Rmax > 0.The supplied passage introduces this result but does not specify the recovered spatial domain.Source: §4 · p. 14

4. Partial-data uniqueness.

With measurements from an open relay-wall subset and radii below Rmax, the data determine the divergence on the union of the corresponding balls. Extending to the full wall and unbounded radii recovers the full-data ambiguity.

4. Partial-data uniqueness.

  • For open Γ and 0 < r < Rmax, equal data are equivalent to equal divergences on ΩΓ,Rmax, the union of the corresponding balls.The converse also holds: divergence agreement there gives equal data on the observation set.Source: §4 · p. 14
  • The proof establishes local uniqueness by covering ΩΓ,Rmax with the relevant balls, yielding divergence equality almost everywhere there.Source: §4 · p. 14
  • Taking Γ = R² and allowing Rmax to grow without bound recovers the full-data ambiguity characterized in Corollary 3.3.Source: §4 · p. 14

5. Algorithm and experiments.

An FFT reconstruction with Stolt interpolation recovers fields from model data, and tests measure its accuracy and invariance to divergence-free perturbations. Imaging examples span model-generated, rendered, and measured transients, with transport mismatch limiting interpretation of rendered and measured vector components.

5.1. Algorithm.

  • The FFTs dominate runtime and memory, giving O(N^3 log N) time and O(N^3) memory on an N^3 grid.Stolt interpolation follows an odd radial extension and FFT; the cone factor is applied before interpolation, with a local even quadratic fit near the cone edge.Source: §5.1 · p. 15

5.2. Validation on test fields.

  • 1.89% and 6.23% are the relative vector errors for the cancelling and noncancelling test fields, respectively, but their differing geometry and depth prevent isolating cancellation’s effect.Source: §5.2 · p. 15
  • Figure 2 shows the axial component is even about the target center with depth cancellation and odd without it; its rows share color scales.Source: §5.2 · p. 16, Figure 2 · p. 16
  • 4.17×10^-8 and 2.88 × 10^-6 are the relative discrete ℓ2 differences in data and reconstructed fields after adding a divergence-free perturbation.Source: §5.2 · p. 16

5.3. Three imaging examples.

  • Rendered and measured transients include transport effects absent from the model, so their displayed vector components are not interpreted as surface-normal estimates.Background correction, Savitzky–Golay differentiation, and Gaussian smoothing stabilize finite-window and noisy data; these operations are not part of the continuous inversion formula.Source: §5.3 · p. 16
  • The Letter-T reconstruction reproduces the reference’s horizontal bar and vertical stem, with slight edge blurring.The reconstruction uses model-generated data; the figure presents reference and reconstructed potential alongside field magnitude and normalized components.Source: §5.3 · p. 17, Figure 3 · p. 17
  • Rendered Bunny images retain the ears and body outline, with interruptions near the head and uneven response across the body.Source: §5.3 · p. 17
  • Measured statue images show the extended arm, bent torso, and base; field magnitude is more localized than the diffuse-halo potential, while fine detail remains unresolved.Source: §5.3 · p. 17

6. Conclusion.

Partial measurements determine the divergence in the corresponding observable region. The numerical experiments illustrate reconstructions, but finite-aperture, radial-truncation, and raw-noise effects remain unquantified.

6. Conclusion.

  • The Bunny panels show the reference scene, reconstructed potential, field magnitude, and normalized field components.Source: Figure 4 · p. 18
  • Partial measurements retain the scalar structure and determine the divergence in the corresponding observable region.Source: §6 · p. 18
  • Finite apertures, radial truncation, and raw-data noise are not quantified; nonconfocal reconstruction is left for future work.Source: §6 · p. 18

Appendix A. Proofs for the forward operator and Fourier–sine representation.

Appendix A supplies estimates for the forward operator and establishes the Fourier representation used in the reconstruction. It also verifies boundedness and localization properties of the transformed data.

Appendix A. Proofs for the forward operator and Fourier–sine representation.

  • The proof bounds the area of wall centers whose radius-r balls contain a fixed point by πr^2, yielding the forward-operator estimate.Source: Appendix A · p. 18
  • If the field is supported at depths beyond d, data vanish for radii r < d because the relay-plane-centered balls lie within the empty slab.Source: Appendix A · p. 18
  • The preprocessed forward data have polynomial L2 growth in radius and define a tempered distribution.The estimate also gives continuity from L2(R3) to L2(R2 × (−R, R)) for every R > 0.Source: Appendix A · p. 19
  • The Fourier argument establishes a representation of the transformed preprocessed data by Gh.The proof verifies local integrability and uses the change of variables ρ = |k| to obtain the representation.Source: Appendix A · p. 21

Appendix B. Supporting estimates and constructions.

A constructed depth-cancelling irrotational Schwartz family shows that the standard Sobolev data norm cannot be uniformly bounded by the reconstruction norm.

Appendix B. Supporting estimates and constructions.

  • A depth-cancelling irrotational Schwartz family demonstrates that the standard Sobolev data norm cannot be bounded uniformly by the reconstruction norm.Source: Appendix B · p. 21

B.1. A depth-cancelling family near tangential frequencies.

A depth-cancelling family with spectral support near tangential frequencies exposes a mismatch between standard data regularity and the reconstruction norm. As the vertical frequency approaches zero, the ratio of these norms diverges.

B.1. A depth-cancelling family near tangential frequencies.

  • The construction uses a spectrum supported where 1 < |k′| < 2, separated from the origin to give a smooth compactly supported scalar spectrum.Source: §B.1 · p. 21
  • The family has zero depth integral, and its vertical spectral support lies in ε ≤ |k3| ≤ 2ε while 1 ≤ |k| ≤ √Source: §B.1 · p. 22
  • As ε ↓ 0, the standard data Sobolev norm divided by the reconstruction norm diverges for the constructed family.Source: §B.1 · p. 22
  • No constant Cs provides the stated uniform Sobolev bound throughout the depth-cancelling Schwartz class.Source: §B.1 · p. 22

B.2. Spherical means at the radial origin.

The radial lift is represented by the spherical-mean transform used in. If Rh vanishes on V × (0, R), the transform vanishes on V × B_R^3(0), yielding the asserted local L2 regularity.

B.2. Spherical means at the radial origin.

  • The radial lift G agrees with the spherical-mean transform used in.Source: §B.2 · pp. 22–23
  • Rh = 0 on V × (0, R) implies the transform vanishes on V × B_R^3(0), yielding the asserted local L2 regularity.Source: §B.2 · pp. 22–23
  • Approximating h by x3-even Schwartz functions and applying the same estimate extends the transform representation to h.The transforms converge by continuity, while the radial lifts converge locally in L2.Source: §B.2 · p. 23
Loading 2609.26846v1…