A Low-Dimensional Representation for Robust Partial Isometric Correspondences Computation

Alan Brunton, Michael Wand, Stefanie Wuhrer, Tino Weinkauf, Hans-Peter Seidel

Abstract

Intrinsic isometric shape matching has become the standard approach for pose invariant correspondence estimation among deformable shapes. Most existing approaches assume global consistency, i.e., the metric structure of the whole manifold must not change significantly. While global isometric matching is well understood, only a few heuristic solutions are known for partial matching. Partial matching is particularly important for robustness to topological noise (incomplete data and contacts), which is a common problem in real-world 3D scanner data. In this paper, we introduce a new approach to partial, intrinsic isometric matching. Our method is based on the observation that isometries are fully determined by purely local information: a map of a single point and its tangent space fixes an isometry for both global and the partial maps. From this idea, we develop a new representation for partial isometric maps based on equivalence classes of correspondences between pairs of points and their tangent spaces. From this, we derive a local propagation algorithm that find such mappings efficiently. In contrast to previous heuristics based on RANSAC or expectation maximization, our method is based on a simple and sound theoretical model and fully deterministic. We apply our approach to register partial point clouds and compare it to the state-of-the-art methods, where we obtain significant improvements over global methods for real-world data and stronger guarantees than previous heuristic partial matching algorithms.

Introduction

In computer graphics and vision, shape matching involves identifying corresponding points between two 3D surfaces. While rigid matching is well-established, matching deformable objects—such as a human in different poses—remains a significant challenge. Most state-of-the-art methods rely on global isometry, assuming that the intrinsic distance (measured along the surface) between any two points remains constant.

However, real-world 3D scans are rarely perfect. They often suffer from "topological noise," such as holes from occlusions or "bridges" created when two parts of a shape touch (e.g., a hand resting on a hip). These artifacts distort global geodesic distances, causing standard global matching algorithms to fail. This research introduces a robust framework for partial matching, where only specific regions of the surfaces are expected to match isometrically, effectively ignoring corrupted or missing data.

Methodology: A Local-to-Global Approach

The core innovation of this work is the shift from global invariants to local propagation. The authors leverage a fundamental principle from differential geometry: an isometry between surfaces is uniquely determined by the mapping of a single point and its tangent space (direction). This reduces the search space for a match to just three degrees of freedom (3-DOF): two for the point's position and one for the rotation of the tangent space.

The proposed pipeline: from sparse features to oriented point matches, partial region growing, and final merged result.

The Correspondence Pipeline

As shown in the pipeline above, the process begins by identifying sparse feature points. These are used to generate a set of candidate oriented point matches. Each match acts as a "seed" for a local map.

Isometric Region Growing

Instead of solving for the entire shape simultaneously, the algorithm employs a "flood-fill" approach. Starting from a seed match, the correspondence is propagated to neighboring points using parallel transport.

The iterative process of isometric region growing starting from a single oriented point match.

During this growth phase, the algorithm monitors a stretch constraint. If the local geometry begins to deviate significantly—typically at the boundary of a hole or a contact point—the growth stops. This ensures that the resulting partial maps are locally consistent and undistorted by topological noise.

Agglomerative Clustering

Because many different seeds may represent the same underlying partial match, the final step involves a manifold-based clustering technique. This groups individual partial maps into a single, consistent, and robust correspondence, filtering out false matches caused by noise.

Experimental Results

The method was evaluated on several challenging datasets, including the Samba sequence (a woman dancing in a dress) and the SCAPE dataset.

  • Robustness to Contacts: In the Samba sequence, where the dress frequently touches the legs, global methods like Blended Intrinsic Maps (BIM) often produce misalignments. This local propagation method correctly identifies these as separate parts, maintaining a semantically accurate mapping.
  • Handling Missing Data: The algorithm naturally "flows" around holes in the mesh, preventing the error from missing data from polluting the rest of the calculation.
  • Quantitative Performance: On the SCAPE benchmark, the method achieved lower geodesic error than state-of-the-art global methods in regions where isometry was preserved.

Discussion and Limitations

The primary advantage of this approach is its robustness; by treating redundancy as a filter, the algorithm can distinguish true isometric matches from noise-induced artifacts. However, the method is computationally intensive, with full correspondences sometimes taking several hours to compute. Additionally, it remains dependent on the initial identification of at least a few accurate feature "seeds." Future work may involve combining this local approach with continuous symmetry detection to improve speed and further prune the search space.

Summary updated

Abstract

Intrinsic isometric shape matching has become the standard approach for pose invariant correspondence estimation among deformable shapes. Most existing approaches assume global consistency, i.e., the metric structure of the whole manifold must not change significantly. While global isometric matching is well understood, only a few heuristic solutions are known for partial matching. Partial matching is particularly important for robustness to topological noise (incomplete data and contacts), which is a common problem in real-world 3D scanner data. In this paper, we introduce a new approach to partial, intrinsic isometric matching. Our method is based on the observation that isometries are fully determined by purely local information: a map of a single point and its tangent space fixes an isometry for both global and the partial maps. From this idea, we develop a new representation for partial isometric maps based on equivalence classes of correspondences between pairs of points and their tangent spaces. From this, we derive a local propagation algorithm that find such mappings efficiently. In contrast to previous heuristics based on RANSAC or expectation maximization, our method is based on a simple and sound theoretical model and fully deterministic. We apply our approach to register partial point clouds and compare it to the state-of-the-art methods, where we obtain significant improvements over global methods for real-world data and stronger guarantees than previous heuristic partial matching algorithms.
BibTeX
@inproceedings{brunton2014low,
  author = {Alan Brunton and Michael Wand and Stefanie Wuhrer and Tino Weinkauf and Hans-Peter Seidel},
  title = {{A Low-Dimensional Representation for Robust Partial Isometric Correspondences Computation}},
  booktitle = {Graph. Model},
  year = {2014},
  url = {https://openreview.net/forum?id=BzuHry5jT8_},
}
Figure