Enhancing 3D Face Recognition: Achieving Significant Gains via 2D-Aided Generative Augmentation

Area of Science:

• Computer Vision
• Biometrics
• Machine Learning

Background:

• Deep learning-based 3D face recognition is limited by scarce, costly 3D facial datasets.
• Acquiring large-scale 3D facial data is labor-intensive and expensive.
• Existing methods struggle with data availability for robust 3D face recognition.

Purpose of the Study:

• To develop a scalable and cost-effective data augmentation method for 3D face recognition.
• To enable 3D face reconstruction from abundant 2D images for training.
• To improve the performance of deep learning models in 3D face recognition using synthetic data.

Main Methods:

• A novel 2D-aided framework for 3D face reconstruction from 2D images.
• Integration of 3D face reconstruction with normal component image encoding.
• Fine-tuning deep face recognition models using synthetically generated 3D face data.

Main Results:

• Achieved competitive rank-1 accuracies on four public benchmarks (BU-3DFE, FRGC v2, Bosphorus, BU-4DFE).
• Demonstrated high performance (e.g., 99.2% on BU-3DFE) without real 3D training data.
• Confirmed that higher-fidelity 3D inputs improve recognition accuracy.

Conclusions:

• The proposed framework offers an efficient and scalable paradigm for 3D face recognition.
• Leveraging 2D images for synthetic 3D data is effective for practical biometric systems.
• This approach provides insights into data-efficient training strategies for recognition tasks.