Learning Low-Rank Class-Specific Dictionary and Sparse Intra-Class Variant Dictionary for Face Recognition.

Area of Science:

• Computer Vision
• Machine Learning
• Pattern Recognition

Background:

• Face recognition is complex due to variations in illumination, expression, and occlusion.
• Sparse Representation based Classification (SRC) performs well with sufficient training data but struggles with the small sample size problem.

Purpose of the Study:

• To present a novel face recognition framework (LRSE+SC) that addresses the small sample size problem and variations in facial images.
• To improve face recognition accuracy in challenging conditions with limited training data.

Main Methods:

• Utilizing low-rank and sparse error matrix decomposition (LRSE) to separate discriminative features from intra-class variations.
• Employing sparse coding techniques (SC) with a supervised dictionary for representative features and a shared within-individual dictionary for variations.
• Adopting a reconstruction-based scheme for final classification.

Main Results:

• The proposed LRSE+SC method demonstrates robustness against illumination, expression variations, and occlusions.
• Achieved state-of-the-art performance on benchmark datasets including AR, FERET, FRGC, and LFW.
• Effectively handles corrupted training data and scenarios with insufficient samples per subject.

Conclusions:

• LRSE+SC offers a significant advancement in face recognition, particularly for scenarios with limited training data.
• The framework's ability to model intra-class variations via a shared dictionary enhances recognition accuracy.
• The method provides a robust solution for real-world face recognition challenges.