Undersampled face recognition via robust auxiliary dictionary learning.

Area of Science:

• Computer Vision
• Machine Learning
• Biometrics

Background:

• Face recognition systems often struggle with limited training data per subject.
• Intraclass variations (illumination, expression) and occlusions pose significant challenges.
• Existing methods may not adequately address unseen corruptions during training.

Purpose of the Study:

• To develop a robust face recognition approach for scenarios with undersampled training data.
• To enhance recognition accuracy despite large intraclass variations and un-seen occlusions.
• To outperform current state-of-the-art sparse representation-based methods.

Main Methods:

• Learning a robust auxiliary dictionary from subjects not of interest.
• Utilizing undersampled training data to handle both intra- and inter-class variations.
• Developing a method to automatically disregard unseen occlusions.

Main Results:

• The proposed method effectively handles large intraclass variations like illumination and expression.
• It demonstrates robustness against occlusions and disguises not present during training.
• Experimental results on four datasets confirm superior performance over state-of-the-art methods.

Conclusions:

• The novel approach offers effective and robust face recognition with limited training data.
• It successfully addresses unseen occlusions by learning from auxiliary data.
• This method represents a significant advancement in handling challenging face recognition scenarios.