Shape Alignment via Allen-Cahn Nonlinear-Convection

Daniel Solano, Jerome Darbon, Laurent Younes · 2025

Read Shape Alignment via Allen-Cahn Nonlinear-Convection by Daniel Solano, Jerome Darbon, Laurent Younes (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-16281.txt.

Introduction

Daniel Solano, Jerome Darbon, and Laurent Younes aligned shapes of differing topology with a phase field method in a 2025 paper. Comparing shapes that may merge or split breaks classic registration, which assumes matching structure. Their control problem asks for the cheapest flow that morphs one binary image into another, within a small error.

The morphing follows a generalized Allen-Cahn flow with convection, steered by a normal speed and a smooth velocity field. The team proved mild solutions exist and characterized the minima of the time-stepped problem. Simulations of the fully discrete version respected the expected bounds in practice. The minimal control cost becomes a new distance between shapes, aimed at anatomy and image data work. Stronger solution theory is left as future work.

Catalog record

CREATORDaniel Solano, Jerome Darbon, Laurent Younes
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE64 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2504-16281

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Shape Alignment via Allen-Cahn Nonlinear-Convection Daniel Solano, Jerome Darbon, Laurent Younes Abstract This paper demonstrates the impact of a phase field method on shape registration to align shapes of possibly different topology. It yields new insights into the building of discrepancy measures between shapes regardless of topology, which would have applications in fields of image data analysis such as computational anatomy. A soft end-point optimal control problem is introduced whose minimum measures the minimal control norm required to align an initial shape to a final shape, up to a small error term. The initial data is spatially integrable, the paths in control spaces are integrable and the evolution equation is a generalized convective Allen-Cahn. Binary images are used to represent shapes for the initial data. Inspired by level-set methods and large diffeomorphic deformation metric mapping, the controls spaces are integrable scalar functions to serve as a normal velocity and smooth reproducing kernel Hilbert spaces to serve as velocity vector fields. The existence of mild solutions to the evolution equation is proved, the minimums of the time discretized optimal control problem are characterized, and numerical simulations of minimums to the fully discretized optimal control problem are displayed. The numerical implementation enforces the maximum-bounded principle, although it is not proved for these mild solutions. This research offers a novel discrepancy measure that provides valuable ways to analyze diverse image data sets. Future work involves proving the existence of minimums, existence and uniqueness of strong solutions and the maximum bounded principle. Keywords: Phase-field Model; Large Diffeomorphic Deformation Metric Mapping; Level Set Methods, Allen-Cahn Equation; Optimal Control; Bounded Maximum Principle 1 Introduction In his seminal treatise “On Growth and Forms” [ 53 ] , D’Arcy Thompson introduced the fundamental principle of analyzing the similarities between two shapes through the construction of transformations mapping one of them into the other. This principle was later formalized in Grenander’s pattern theory [ 30 , 29 ] , the theory developing into practical tools with the introduction of diffeomorphic registration [ 66 ] and, among others, the large deformation diffeomorphic mapping algorithm [ 36 , 11 , 54 , 55 , 20 , 7 ] , or LDDMM. This approach, along with many others targeting diffeomorphic registration [ 6 , 24 , 17 , 59 , 5 ] , has been used for applications that extract information from shape changes, including, in particular, medical studies based on computational anatomy [ 31 ] . These methods respect the basic paradigm of D’Arcy Thompson theory of transformations in that they evaluate, for comparison purposes, the “simplest” diffeomorphic transformation mapping one shape onto another. Restricting to such transformations has several advantages, one of the foremost being the structure of the diffeomorphism groups and the existence of well-studied Riemannian metrics [ 4 , 40 ] , with transformations conveniently represented as flows of ordinary differential equations [ 16 , 55 , 66 ] . In addition, diffeomorphisms are easily interpretable, as they can register target shapes to a fixed “template” allowing for the transport of existing information (such as region labeling) from the latter to the former. Moreover, morphometric measures based on the deformation Jacobian are easily computable and can be used in further analysis [ 5 ] . Shape spaces represented as sets of diffeomorphic transformations of a template can be build to include a wild variety of objects, albeit all within the same topological class. Eliminating this topological constraint is challenging, however, as it may require to relinquish many of the properties that were just listed. Non-diffeomorphic comparison between images have been introduced in the context of metamorphosis [ 41 , 57 , 33 , 47 , 46 ] , which estimates transformations between images by optimizing displacements both in space and image intensity. There is no canonical modification of this approach to comparing shapes, defined as curves or surfaces, however, in that running metamorphosis between two images with given zero level, would depend on the choice made for this representation, and not on the level sets themselves. This limitation is partially addressed in [ 34 ] , in which shapes are embedded in a much larger space of varifolds, which can be described, in codimension one, as products of measures on \mathbb{R}^{d} (the varifold weight) and transition probabilities from \mathbb{R}^{d} to the unit sphere. The metamorphosis mechanism is then applied to the weight component of the varifold. Since not all varifolds correspond to geometric curves or surfaces, this representation does not define, however, shape trajectories. Other work addressing the partial matching problem also include [ 3 , 51 ] . Of course, the ability to numerically generate topological changes by tracking the level sets of a function satisfying partial differential equations has a long history [ 44 , 43 , 45 , 48 ] , but such processes, which have found a wide range of applications in shape regularization, using, in particular, mean-curvature motion [ 25 , 28 , 22 ] , and in shape segmentation [ 14 , 60 , 15 , 39 ] , have had limited impact in shape analysis, mainly because these equations simplify their solutions (and the associated level sets) in a non-reversible way, which limits their usefulness in transporting spatial information. In this paper, we provide a new strategy for shape analysis, in which a template shape is controlled by a combination of a diffeomorphic and normal flows while being submitted to a small amount of mean-curvature motion, in order to align with a target shape that is only subject to the mean-curvature motion. Shapes being described as two- or three-dimensional sets with finite Lebesgue measure, their evolution is described using a nonlinear parabolic partial differential evolution equation of their characteristic functions We prove the global existence and uniqueness of this parabolic equation—specifically a generalized convective Allen Cahn equation. We set up a time discretization of an optimal control problem with running and terminal cost and prove a Pontryagin’s maximum principle for this problem. And finally, we show some numerical results in which we transform the characteristic function of an initial shape into the mollified characteristic function of a target shape with possibly different topology. Our critical contribution is in designing a meaningful discrepancy measure between shapes that is based on the minimal energy spent transforming one shape into another shape, regardless of whether they share the same topology. Section 2.3 provides the statement of our main theorem, preceded by the necessary definitions and background results, while Section 3 sets up the optimal control problem implemented in our experiments. The main theorem is proved in Section 4 . Section 5 describes the discrete-time implementation of the method as an optimal control problems and make explicit the discrete Pontryagin maximum principle providing necessary conditions satisfied by solutions. Finally, Section 6 provides experimental illustrations of the method. 2 Main theorem 2.1 Summary of notation If B is a Banach space and r\geq 1 , L^{r}(\Omega,B) is the space of r -integrable functions (in the Bochner sense) from a measured space \Omega to B with the usual L^{r} norm that we will denote \|\ \|_{L^{r}} . The notation is reduced to L^{r}(\Omega) when \Omega=\mathbb{R} . We will also only write L^{r} when \Omega and B are

Read the complete text