3D shape isometric correspondence by spectral assignment.

Area of Science:

• Computer Vision
• Computer Graphics
• Computational Geometry

Background:

• Establishing correspondences between 3D shapes is a fundamental problem in computer vision and graphics.
• Existing methods often struggle with variations in pose and require complex feature engineering.

Purpose of the Study:

• To propose a general framework for building 3D shape correspondences by leveraging the isometric property.
• To demonstrate that this problem can be reduced to spectral assignment.

Main Methods:

• Utilizes the isometric property to establish shape correspondences.
• Reduces the problem to spectral assignment by computing the principal eigenvector of a pairwise correspondence matrix.
• Employs local shape features for initial candidate pairs and geodesic distances for matrix construction.

Main Results:

• The proposed framework successfully finds correspondences between 3D shapes.
• The method demonstrates robustness across a variety of poses.
• Achieves significant improvement compared to existing state-of-the-art methods.

Conclusions:

• The isometric property provides a powerful tool for 3D shape correspondence.
• Spectral assignment offers an effective solution for this problem.
• The proposed framework presents a robust and high-performing approach for 3D shape matching.