Detail-Preserving Shape Unfolding.

Area of Science:

• Computer Graphics
• Computational Geometry
• Geometric Modeling

Background:

• Canonical extrinsic representations are crucial for non-rigid shape analysis in computer graphics, enabling applications like shape correspondence and retrieval.
• Existing methods, often based on Multidimensional Scaling (MDS), suffer from significant geometric distortions, limiting their practical utility.
• The need for pose-invariant signatures that are robust to non-rigid transformations is a key challenge in shape analysis.

Purpose of the Study:

• To develop a novel algorithm for 3D shape unfolding that generates pose-invariant canonical representations.
• To effectively preserve the local geometric structure of 3D models during the unfolding process.
• To significantly reduce geometric distortions compared to existing state-of-the-art methods.

Main Methods:

• A novel shape unfolding algorithm is proposed, deforming 3D shapes into a canonical pose invariant to non-rigid transformations.
• The method employs regularization of local rigid transform energy within a shape deformation framework.
• The algorithm involves solving two linear systems iteratively, enhanced by parallel computation and a cascade strategy for robustness.

Main Results:

• The algorithm successfully preserves local structure while minimizing geometric distortion in 3D models.
• Experimental results show superior performance compared to existing state-of-the-art 3D shape unfolding techniques.
• The proposed method demonstrates enhanced efficacy in generating canonical representations for non-rigid shapes.

Conclusions:

• The novel shape unfolding algorithm provides an effective solution for generating pose-invariant representations of non-rigid 3D shapes.
• The method's simplicity, computational efficiency, and robustness make it suitable for various computer graphics applications.
• This work advances the field of 3D shape analysis by offering a distortion-reduced approach to canonical form generation.