Abstract: We propose a method to simultaneously compute scalar basis functions with an associated functional map for a given pair of triangle meshes. Unlike previous techniques that put emphasis on smoothness with respect to the Laplace--Beltrami operator and thus favor low-frequency eigenfunctions, we aim for a spectrum that allows for better feature matching. This change of perspective introduces many degrees of freedom into the problem which we exploit to improve the accuracy of our computed correspondences. To effectively search in this high dimensional space of solutions, we incorporate into our minimization state-of-the-art regularizers. We solve the resulting highly non-linear and non-convex problem using an iterative scheme via the Alternating Direction Method of Multipliers. At each step, our optimization involves simple to solve linear or Sylvester-type equations. In practice, our method performs well in terms of convergence, and we additionally show that it is similar to a provably convergent problem. We show the advantages of our approach by extensively testing it on multiple datasets in a few applications including shape matching, consistent quadrangulation and scalar function transfer.
Introduction
Shape matching—identifying corresponding points between two 3D surfaces—is a fundamental task in computer graphics and geometry processing. The Functional Maps (FM) framework revolutionized this field by shifting the problem from matching millions of points directly to matching functions defined on those surfaces. Traditionally, these functions are represented using a fixed basis, typically the eigenfunctions of the Laplace-Beltrami (LB) operator. While effective for perfectly isometric shapes, these generic geometric bases often fail to capture the high-frequency features necessary for matching non-isometric or semantically similar shapes, such as different people in varying poses.
The FMBS Framework
This paper introduces FMBS (Functional Map and Basis Search), a method that simultaneously optimizes for both the functional map and the underlying basis functions. Instead of relying on a pre-defined geometric basis, FMBS leverages data-driven Proper Orthogonal Decomposition (POD) modes derived from input descriptors like Wave Kernel Signatures.
As illustrated in Figure 2, POD modes are significantly more expressive than standard LB eigenfunctions. While LB modes are great for representing global, smooth geometry, they are independent of the actual features being matched. In contrast, POD modes are "tailored" to the data, capturing localized feature information that is often lost in standard spectral truncation. By allowing the basis to "move" and adapt during optimization, the framework can find a subspace where the shapes align more naturally.
Optimization via ADMM
The resulting joint optimization problem is highly non-linear and non-convex. To solve it efficiently, the authors employ the Alternating Direction Method of Multipliers (ADMM). This approach decomposes the complex problem into simpler, iterative sub-problems:
- Bases Update: Solving Sylvester-type equations to update the basis matrices ($B_1, B_2$).
- Map Update: A simple least-squares step to update the functional map ($C$).
- Constraint Enforcement: Updating dual variables to ensure orthogonality and incorporate regularizers like cycle-consistency and Dirichlet energy.
Despite the increased complexity of searching for both a map and a basis, performing the optimization in a reduced spectral domain (typically around 20 modes) makes the method computationally efficient.
Evaluation and Results
The FMBS method was rigorously tested on several standard benchmarks, including FAUST, SCAPE, and TOSCA. It demonstrates significant improvements in accuracy, particularly for "inter-subject" matching where geometric variations between different individuals are more pronounced.
Figure 5 shows the performance gain on the SCAPE dataset. The "Ours" curve (representing FMBS) consistently achieves a higher percentage of correct correspondences at lower geodesic error thresholds compared to existing state-of-the-art methods like FMAPS and AJD. Beyond simple matching, the improved maps facilitate higher-quality results in downstream applications such as consistent quadrangulation and texture transfer, where it successfully avoids the "bleeding" artifacts common in LB-based methods.
Conclusion
By treating basis selection as a data-driven optimization problem rather than a fixed geometric choice, FMBS provides a more flexible and accurate framework for 3D shape analysis. While the quality of the results depends on the initial descriptors, the method's ability to capture high-frequency features makes it a powerful tool for matching complex, non-isometric shapes in real-world scenarios.
Summary updated