[Paper Review] On Discrete Conformal Seamless Similarity Maps

Marcel Campen, Denis Zorin|arXiv (Cornell University)|May 6, 2017

3D Shape Modeling and Analysis1 references3 citations

TL;DR

This paper provides a rigorous convergence and correctness analysis of an iterative algorithm for computing discrete conformal seamless similarity maps on triangle meshes with prescribed holonomy signatures. By linking the method to Newton's algorithm on angle constraints and showing the system arises from a convex energy, the authors prove convergence under mild conditions and analyze the role of edge flips in maintaining validity during optimization.

ABSTRACT

An algorithm for the computation of global discrete conformal parametrizations with prescribed global holonomy signatures for triangle meshes was recently described in [Campen and Zorin 2017]. In this paper we provide a detailed analysis of convergence and correctness of this algorithm. We generalize and extend ideas of [Springborn et al. 2008] to show a connection of the algorithm to Newton's algorithm applied to solving the system of constraints on angles in the parametric domain, and demonstrate that this system can be obtained as a gradient of a convex energy.

Motivation & Objective

• To rigorously analyze the convergence and correctness of an iterative algorithm for computing discrete conformal seamless similarity maps with prescribed holonomy signatures.
• To establish a connection between the algorithm and Newton's method applied to angle constraints in the parametric domain.
• To demonstrate that the system of constraints can be derived as the gradient of a convex energy function.
• To investigate the role of edge flips in maintaining triangle inequality and preventing degenerate configurations during optimization.
• To address open questions regarding infinite sequences of edge flips, similar to those in discrete Ricci flow.

Proposed method

• The algorithm computes a discrete 0-form φ on a cut mesh M_c that induces a discrete conformal metric via edge length updates l_i = l_i^G * exp((φ_v + φ_w)/2).
• It formulates the holonomy constraints as a system of equations on angles, which are shown to be the gradient of a convex energy function E(ψ).
• The method uses a closed 1-form ξ derived from a basis of harmonic 1-forms, with coefficients ψ related to vertex and cycle angle sums via ψ = Py.
• A modified Newton-type iteration is employed, with line search-like truncation to prevent triangle inequality violations, triggering edge flips when necessary.
• The edge flips are intrinsic and preserve the total angle and geodesic curvature values, maintaining energy and constraint consistency.
• The system is analyzed using a pseudoinverse formulation P+ to relate target constraints (b_t) to the energy gradient, ensuring consistency with holonomy signatures.

Experimental results

Research questions

• RQ1 Does the algorithm converge to a discrete conformal seamless similarity map with the prescribed holonomy signature?
• RQ2 Can the system of angle constraints be interpreted as the gradient of a convex energy function?
• RQ3 What is the impact of edge flips on convergence, and can an infinite sequence of edge flips occur?
• RQ4 Under what conditions can two adjacent triangles degenerate simultaneously, leading to zero-length edges?
• RQ5 How does the algorithm maintain validity when triangle inequality is violated during optimization?

Key findings

• If the algorithm converges, it produces a discrete conformal seamless similarity map with the prescribed holonomy signature.
• The system of constraints on angles is shown to be the gradient of a convex energy function, enabling a Newton-type optimization approach.
• A modified version of the algorithm converges unless an infinite sequence of edge flips occurs, though such a sequence remains unproven to be possible.
• Edge flips are intrinsic and preserve the total angle sums and geodesic curvatures, maintaining consistency with the holonomy constraints.
• The energy function E(ψ) and its modified version E'(ψ) are shown to have gradients that match the left-hand sides of the angle constraint equations.
• The method avoids extending the energy to invalid states by truncating steps before violations and resolving degeneracies via edge flips, following a strategy analogous to Luo (2004).