[Paper Review] A new reconstruction method in integral geometry

Victor Palamodov|arXiv (Cornell University)|Sep 11, 2011

Photoacoustic and Ultrasonic Imaging10 references3 citations

TL;DR

This paper presents a novel analytic inversion method for integral transforms in integral geometry, using a regular generating function Φ(x;λ,φ) = λ + ψ(x,φ) to derive explicit reconstruction formulas. The key contribution is a reconstruction formula for functions in L²(X) with compact support, valid when the principal value integral N(x,y) vanishes for x ≠ y, enabling exact inversion via a generalized Funk-Radon transform in both 2D and higher dimensions.

ABSTRACT

A general method for analytic inversion of geometric integral transforms is proposed

Motivation & Objective

• To develop a general analytic inversion method for geometric integral transforms in integral geometry.
• To address the reconstruction problem of a function f from its integrals over curves defined by a generating function Φ.
• To establish conditions under which exact inversion formulas exist, particularly when the kernel function N(x,y) vanishes.
• To generalize the reconstruction formula to higher dimensions (n ≥ 2) using real analytic generating functions.
• To provide explicit formulas for reconstruction in cases such as α-curves and β-curves with rotational symmetry.

Proposed method

• Proposes a Funk-Radon transform MΦf(σ) defined via integration over level sets F(σ) = {x ∈ X : Φ(x,σ) = 0}, weighted by the gradient of Φ.
• Introduces a principal value integral N(x,y) = Re ∫₀²π dφ / (ψ(x,φ) - ψ(y,φ) ± i0)² for x ≠ y, which must vanish for the reconstruction to hold.
• Derives a reconstruction formula f(x) = −1/(4π²D(x)) ∫₀²π ∫ℝ MΦf(λ,φ)/Φ²(x;λ,φ) dλ dφ, where D(x) is a normalization factor involving the gradient of ψ.
• Extends the method to higher dimensions by defining N(x,y) = Re ∫_{S^{n−1}} dω / (ψ(x,ω) − ψ(y,ω) ± i0)^n and D(x) = 1/|S^{n−1}| ∫_{S^{n−1}} dω / |∇gψ(x,ω)|^n.
• Uses Fourier integral operator techniques and complex analysis to handle the singular integrals, ensuring convergence in L²_loc(X).
• Applies the method to symmetric families such as α-curves (ψ(x,φ) = −r^k cos(kθ − φ)) and β-curves, leveraging Z_k symmetry to simplify the kernel.

Experimental results

Research questions

• RQ1 Under what conditions can an exact analytic inversion formula be derived for integral transforms in integral geometry?
• RQ2 How does the vanishing of the principal value kernel N(x,y) for x ≠ y affect the solvability of the reconstruction problem?
• RQ3 Can the reconstruction formula be generalized from 2D to higher-dimensional manifolds with Riemannian metrics?
• RQ4 What role does the generating function Φ(x;λ,φ) = λ + ψ(x,φ) play in enabling explicit inversion when ψ is real analytic?
• RQ5 How do symmetries (e.g., Z_k) in the curve family simplify the reconstruction process and ensure the kernel N(x,y) vanishes?

Key findings

• The reconstruction formula f(x) = −1/(4π²D(x)) ∫₀²π ∫ℝ MΦf(λ,φ)/Φ²(x;λ,φ) dλ dφ holds for all f ∈ L²(X)_comp when N(x,y) ≡ 0 for x ≠ y.
• The integral converges in L²_loc(X), ensuring stability and applicability in practical reconstruction settings.
• For α-curves with ψ(x,φ) = −r^k cos(kθ − φ), the kernel N(x,y) vanishes due to the presence of exactly two real zeros in ψ(x,φ) − ψ(y,φ), satisfying the key condition.
• In higher dimensions (n ≥ 2), the reconstruction formula generalizes to f(x) = 1/((2πi)^n D(x)) Re ∫Σ MΦf(λ,ω)/(Φ(x;λ,ω)+i0)^n dλ dω for even n, and to a delta-function form for odd n.
• The normalization factor D(x) depends only on the conformal class of the Riemannian metric g, making the formula invariant under conformal changes.
• The method provides exact inversion, unlike Beylkin’s parametrix approach, which only gives high-frequency approximations.