Related Experiment Video
Updated: Apr 4, 2026
06:26
Quantitative Assessment Protocol for Facial Soft Tissue Volumetric Changes with Stereophotogrammetry
Published on: December 9, 2025
294
A Two-Stage Framework for 3D Face Reconstruction from RGBD Images.
IEEE Transactions on Pattern Analysis and Machine Intelligence
|September 10, 2015
Summary
This study presents a novel two-stage algorithm for accurate 3D face reconstruction from low-quality RGBD images. The method effectively handles noise and variations in pose and expression, creating realistic face models.
Area of Science:
- Computer Vision
- 3D Graphics
- Machine Learning
Background:
- 3D face reconstruction from RGBD images is challenging due to noise and variability.
- Existing methods struggle with low-quality depth data and diverse facial expressions.
Purpose of the Study:
- To develop a robust 3D face reconstruction method using inexpensive RGBD sensors.
- To address noise and variability in depth maps for realistic face modeling.
Main Methods:
- A novel two-stage algorithm utilizing data-driven local sparse coding and template-based surface refinement.
- Stage one extracts errors from depth patches; stage two smooths boundaries and refines global shape.
Main Results:
- High-resolution and accurate 3D face models are generated from low-quality depth maps.
- The approach demonstrates robustness against substantial random noise and corruption.
- Effective handling of high variability in pose and facial expression.
Conclusions:
- The proposed two-stage algorithm successfully reconstructs realistic 3D faces from noisy, low-resolution RGBD data.
- This method offers a marker-free and user-interaction-free solution for 3D face modeling.
- The approach achieves high accuracy across various viewpoints and expressions.
