Abstract
Traditional surface modeling techniques often struggle with large deformations, frequently producing unnatural geometric artifacts like spikes or wiggles. This paper presents a novel "rotation-strain" representation for surface modeling that leverages plasticity to handle significant shape changes smoothly. By decoupling extrinsic bending (rotations) from intrinsic stretching (strains) and ensuring geometric compatibility via the Gauss-Codazzi equations, the proposed method achieves high-quality results without the numerical instabilities common in purely elastic frameworks.
Introduction
In computer graphics, deforming a 3D mesh to satisfy user-defined landmarks is a fundamental task. Most existing approaches, such as As-Rigid-As-Possible (ARAP) or ShapeOp, are essentially elastic: they penalize deviations from an initial "rest" shape. While effective for small movements, these methods often fail under large strains. When a surface is stretched significantly, the algorithm's attempt to preserve the original dimensions can lead to sharp, unnatural spikes. Furthermore, surfaces are 2D manifolds embedded in 3D space, meaning local deformations must satisfy strict compatibility conditions to form a coherent global geometry.
Methods
The authors introduce a plasticity-based approach that modifies the rest state directly rather than resisting deformation. The core of the method is a new differential representation:
- Rotation-Strain Representation: Local deformation is modeled using a $3 \times 3$ rotation matrix ($R$) for extrinsic orientation (bending) and a $2 \times 2$ symmetric matrix ($S$) for intrinsic in-plane stretching.
- Guaranteed Compatibility: This formulation inherently satisfies the Gauss-Codazzi equations, ensuring that any chosen $R$ and $S$ fields correspond to a valid physical surface.
- Smoothness Optimization: The system solves a global optimization problem to find the smoothest fields of $R$ and $S$ that meet user-defined landmark constraints.
- Discrete Shell Model: Built on a thin-shell simulation framework, the method handles both stretching and bending energy in a physically plausible manner.
Results
The method was evaluated on various geometric meshes, including the Stanford Bunny, a horse, and complex medical muscle templates.
- Visual Quality: Unlike ShapeOp or ARAP, which produce jagged artifacts during extreme stretching, the proposed method maintains perfectly smooth surfaces.
- Performance: In muscle-matching tasks, the surface-based approach was 6x faster per iteration and 3x faster overall compared to volumetric plasticity methods, due to the reduced degrees of freedom in surface meshes.
- Robustness: The framework handles anisotropic stretching and large rotations simultaneously without requiring manual parameter tuning.
Conclusion
While the current implementation is not yet real-time due to the complexity of the global optimization, it represents a significant step forward in robust surface modeling. By treating plasticity as a design tool rather than just a physical simulation effect, the authors provide a way to decouple bending and stretching effectively. Future work includes exploring multigrid solvers for interactive performance and extending the framework to other manifold types, such as 1D curves.
Summary updated