Abstract
This paper studies the space of $BV^2$ planar curves endowed with the $BV^2$ Finsler metric over its tangent space of displacement vector fields. Such a space is of interest for applications in image processing and computer vision because it enables piecewise regular curves that undergo piecewise regular deformations, such as articulations. The main contribution of this paper is the proof of the existence of the shortest path between any two $BV^2$-curves for this Finsler metric. Such a result is proved by applying the direct method of calculus of variation to minimize the geodesic energy. This method applies more generally to similar cases such as the space of curves with $H^k$ metrics for $k \geq 2$ integer. This space has a strong Riemannian structure and is geodesically complete. Thus, our result shows that the exponential map is surjective, which is complementary to geodesic completeness in infinite dimensions. We propose a finite element discretization of the minimal geodesic problem, and use a gradient descent method to compute a stationary point of the energy. Numerical illustrations show the qualitative difference between $BV^2$ and $H^2$ geodesics.
Introduction
In computer vision and medical imaging, "shape registration" is the process of finding an optimal deformation that transforms one object into another. Mathematically, the space of all possible curves is treated as a manifold, and the goal is to find the geodesic—the path of least energy—between two points (shapes) in that space. While standard approaches use Sobolev ($H^k$) Riemannian metrics, these are often too "stiff," forcing curves to be extremely smooth. This stiffness is problematic when dealing with real-world objects that possess sharp corners or articulated joints, such as a human arm bending at the elbow.
Theoretical Framework: $BV^2$ vs. Sobolev Metrics
The authors address this by exploring $BV^2$ (Bounded Variation of the second order) metrics, which are Finslerian in nature. Unlike Riemannian metrics, $BV^2$ allows for piecewise-regular deformations, meaning it can better preserve sharp features and represent mechanical-like movements by allowing "jumps" in the second derivative.
A primary contribution of this work is the rigorous proof of the existence of a minimizing geodesic between any two curves in both $BV^2$ and $H^k$ spaces. Proving existence in infinite-dimensional spaces is non-trivial; the authors employ the direct method of the calculus of variations. To handle the lack of certain compactness properties in $BV^2$ spaces, they utilize a "martingale" strategy, decomposing the path into a sequence of dyadic intervals to ensure convergence.
Numerical Implementation and Results
For practical application, the authors propose a finite element discretization where curves are represented as piecewise affine segments. The path through time is discretized, and the energy is minimized using a gradient descent algorithm. Because the $BV^2$ norm involves non-differentiable $L^1$-like terms, a small smoothing parameter ($\epsilon$) is applied to make the functional differentiable.
Numerical illustrations using synthetic models demonstrate the qualitative differences between metrics. While $H^2$ geodesics tend to "round off" shapes during transformation, $BV^2$ geodesics maintain straight edges and sharp corners. For example, when deforming a square, the $BV^2$ path preserves the flat edges better than the $H^2$ path, which causes the edges to bow and the corners to soften. Furthermore, for Sobolev metrics, the existence of a minimal geodesic implies that the exponential map is surjective, a strong geometric result indicating that these metrics can reach any point in the curve space.
Discussion and Conclusion
Despite the theoretical and qualitative successes, challenges remain. The energy functional is highly non-convex, meaning gradient descent can get stuck in local minima depending on the initial correspondence between curves. Future research may focus on designing Finsler metrics that explicitly favor piecewise-rigid motions, ideal for tracking articulated robots or human skeletons. Overall, this work provides a solid mathematical foundation for using $BV^2$ metrics in shape analysis, offering a more realistic framework for modeling complex deformations.
Summary updated