Histogram-Based CRC for 3D-Aided Pose-Invariant Face Recognition.

Area of Science:

• Computer Vision
• Machine Learning
• Biometrics

Background:

• Traditional Collaborative Representation-based Classification (CRC) methods for face recognition struggle with data uncertainty from pose and illumination variations.
• Existing methods often rely on distance-based metrics, which can be suboptimal for complex variations.

Purpose of the Study:

• To develop a robust pose-invariant face classification method.
• To enhance Collaborative Representation-based Classification (CRC) by incorporating histogram statistical measurements and 3D morphable models (3DMM).

Main Methods:

• A novel histogram-based CRC (H-CRC) method is proposed, integrating a 3D morphable model (3DMM).
• 3DMM is fitted to reconstruct 3D shapes and textures, enabling rendering of frontalized virtual samples from arbitrary poses.
• Histogram information from rendered images is utilized for metric learning and neighbor evaluation, forming a unified 3D-aided CRC framework.

Main Results:

• The proposed H-CRC method demonstrates effective pose-invariant face classification.
• The approach achieves desirable classification results across multiple benchmark face databases (ORL, Georgia Tech, FERET, FRGC, PIE, LFW).

Conclusions:

• The integration of 3DMM and histogram-based metric learning significantly improves face recognition robustness against pose variations.
• The developed 3D-aided CRC framework offers a promising solution for challenging face recognition scenarios.