Abstract:We propose a general convex optimization problem for computing regularized geodesic distances. We show that under mild conditions on the regularizer the problem is well posed. We propose three different regularizers and provide analytical solutions in special cases, as well as corresponding efficient optimization algorithms. Additionally, we show how to generalize the approach to the all pairs case by formulating the problem on the product manifold, which leads to symmetric distances. Our regularized distances compare favorably to existing methods, in terms of robustness and ease of calibration.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is: in shape analysis, how to calculate the regularized geodesic distances to overcome the limitations of traditional geodesic distances in some applications. Specifically:
1. **Smoothness and other properties**: Traditional geodesic distances may not have the required smoothness or other properties (such as alignment to a certain vector field, boundary invariance, etc.). These problems will affect performance in some geometric processing tasks.
2. **Lack of a unified framework**: Although there are multiple methods for calculating geodesic distances and their variants, there is a lack of a unified framework to systematically introduce and control regularization, making these methods difficult to adjust and apply.
To solve the above problems, the author proposes a general framework based on convex optimization for calculating regularized geodesic distances. This framework has the following characteristics:
- **Theoretical guarantee**: Under certain conditions, this optimization problem has a unique solution, and as the regularization weight approaches zero, the solution will converge to the exact geodesic distance.
- **Multiple regularization terms**: Several different regularization terms are provided, and the corresponding analytical solutions and efficient optimization algorithms are given.
- **Extension to all point pairs**: By defining the problem on the product manifold, it can be generalized to the case of all point pairs, thus obtaining a symmetric distance metric.
### Formula summary
The core of this framework is a convex optimization problem, and its general form is:
\[
\text{Minimize} \quad \alpha E(u) - \int_{M} u(x) \, d\text{Vol}(x)
\]
\[
\text{subject to} \quad |\nabla u(x)| \leq 1 \quad \forall x \in M \setminus E
\]
\[
u(x) \leq 0 \quad \forall x \in E
\]
where \(E(u)\) is a regularization term, and its general form is:
\[
E(u) = \int_{M} F(\nabla u(x), x) \, d\text{Vol}(x)
\]
Specific regularization terms include but are not limited to:
- **Dirichlet regularization**:
\[
E_{\text{Dir}}(u) = \frac{1}{2} \int_{M} |\nabla u(x)|^2 \, d\text{Vol}(x)
\]
- **Vector - field alignment regularization**:
\[
E(u) = \frac{1}{2} \int_{M} |\nabla u(x)|^2 + \beta \langle V(x), \nabla u(x) \rangle^2 \, d\text{Vol}(x)
\]
- **Hessian regularization**:
\[
E(u) = \frac{1}{2} \int_{M} |\nabla^2 u(x)|^2 \, d\text{Vol}(x)
\]
By introducing these regularization terms, the performance of geodesic distances in specific applications can be improved while maintaining their basic properties.