Abstract
In the fields of computer graphics and computer vision, a great amount of research and analysis has been conducted on expression-carrying face models. How to construct more realistic and effective 3D face models has become an immense challenge to researchers. This paper proposes a parametric 3D face model editing algorithm based on existing 3D face models. The algorithm takes a number of existing 3D face models with different expressions as input, edits the models through model deformation and interpolation, and generates a new 3D face model. The proposed algorithm has been evaluated through extensive experiments. The results suggest that compared with existing editing methods, the proposed method is more effective and efficient in constructing various 3D face models.
Introduction
Creating realistic 3D facial expressions is a cornerstone of computer graphics, yet it remains a "bottleneck" task. Traditional methods generally fall into two camps: reconstruction (using 3D scanners), which is accurate but time-consuming and lacks flexibility, and creative modeling (using tools like Maya), which requires high manual effort.
Technically, existing mesh editing algorithms often rely on differential coordinates (like Laplacian editing). While these are good at preserving local details, they are notoriously sensitive to rotation. In large-scale deformations—such as moving from a neutral face to a wide-mouthed laugh—these methods can cause the mesh to "collapse" or twist unnaturally. Furthermore, selecting and managing control points in 3D space is mathematically complex and computationally expensive.
Methodology
The proposed framework shifts the heavy lifting from 3D to 2D space to simplify complex mesh deformations. The process follows a multi-step pipeline:
- ICP Registration: Multiple 3D face models (source and targets) are aligned using the Iterative Closest Point (ICP) algorithm to establish a one-to-one vertex correspondence.
- Quasi-Conformal Mapping: The 3D mesh is parameterized (flattened) onto a 2D planar disc. This mapping is "quasi-conformal," meaning it preserves local angles and shapes even as the overall dimensions change.
- 2D Editing & Interpolation: Deformation and interpolation are performed on the 2D plane. The Beltrami coefficient guides the iteration, ensuring the mapping between the source and target planes remains smooth and minimizes distortion.
- Model Restoration: Finally, the edited 2D models are projected back into 3D space using a restoration algorithm based on a cotangent Laplace matrix and least-squares optimization.
Key Insights and Results
The core innovation lies in dimensionality reduction. By moving the editing process to a 2D plane, the algorithm bypasses the "rotation distortion" problem entirely. In 2D, the complex rotations of a 3D jaw or cheek become simple planar shifts. The use of quasi-conformal mapping ensures that even if the face is stretched, the "local signature" (the specific curve of a nostril or the fold of an eyelid) remains consistent.
Experimental results demonstrate that the proposed method achieved high-quality results using only 24–26 control points, whereas previous state-of-the-art methods required 135–200 points for similar tasks. Qualitative comparisons showed that while traditional Laplacian methods caused structural collapses during large expression changes, the proposed algorithm maintained structural integrity. The total processing time for model editing and restoration ranged from approximately 16 to 32 seconds, making it efficient for iterative design workflows.
Conclusion
This paper presents a novel 3D face editing framework that simplifies complex 3D mesh deformations by mapping them onto a 2D planar disc using quasi-conformal mapping. While the method significantly reduces computational complexity and rotation distortion, future work will focus on automating the manual selection of characteristic points and exploring deep learning-based registration to handle more extreme expression changes.
Summary updated