Nasal Patches and Curves for Expression-Robust 3D Face Recognition

Area of Science:

• Computer Science
• Biometrics
• Artificial Intelligence

Background:

• 3D face recognition is crucial for security and identification.
• Existing methods often struggle with variations in facial expressions.
• The nasal region's potential for robust recognition remains underexplored.

Purpose of the Study:

• To investigate the efficacy of the nasal region for expression-robust 3D face recognition.
• To develop and evaluate a novel algorithm for nasal region-based 3D face recognition.

Main Methods:

• A five-step algorithm involving coarse nose tip detection, face segmentation, alignment, and nasal region cropping.
• Accurate nasal landmarking to identify seven keypoints.
• Feature extraction using surface normals of Gabor-wavelet filtered depth maps and localized spherical patches/curves.
• A genetic algorithm-based feature selector for stable feature identification across expressions.

Main Results:

• Achieved the highest reported nasal region-based recognition ranks on FRGC, Bosphorus, and BU-3DFE datasets.
• Performance is comparable to, and often better than, state-of-the-art 3D face recognition algorithms using the whole face.
• Demonstrated robustness with a single sample per subject and without complex alignment or denoising.

Conclusions:

• The nasal region is a highly effective area for expression-robust 3D face recognition.
• The proposed algorithm offers a robust, efficient, and training-free approach for 3D face recognition.
• This method presents a promising alternative to traditional whole-face recognition techniques.