Compositional Dictionaries for Domain Adaptive Face Recognition.

Area of Science:

• Computer Vision
• Machine Learning
• Pattern Recognition

Background:

• Face recognition systems struggle with variations in pose, illumination, and resolution.
• Developing robust methods to handle these domain shifts is crucial for practical applications.

Purpose of the Study:

• To develop a dictionary learning approach for domain-invariant sparse coding in face recognition.
• To enable accurate face recognition across different viewpoints, illumination conditions, and resolutions.

Main Methods:

• Learning a domain base dictionary and using sparse representations for domain shifts (identity, pose, illumination).
• Adapting dictionaries through sparse linear combinations of the base dictionary.
• Decomposing face images into sparse representations for subject, pose, and illumination.

Main Results:

• Achieved pose and illumination insensitive face recognition through consistent sparse subject representations.
• Demonstrated the ability to estimate pose and illumination conditions from sparse representations.
• Showcased pose alignment and illumination normalization by composing sparse representations.

Conclusions:

• The proposed compositional dictionary approach effectively addresses domain variations in face recognition.
• The method enables robust face recognition and provides capabilities for pose/illumination estimation and normalization.
• Experimental results on public datasets validate the effectiveness of the approach.