Front-to-Side Hard and Soft Biometrics for Augmented Zero-Shot Side Face Recognition

Area of Science:

• Computer Science
• Biometrics
• Artificial Intelligence

Background:

• Face recognition technology is crucial for individual identification.
• Existing research predominantly focuses on frontal face views, neglecting side-face perspectives.
• Side-face views are vital for identity, often being the sole available biometric data.

Purpose of the Study:

• To develop and evaluate new soft biometric traits for invariant extraction from both front and side faces.
• To augment zero-shot side face recognition by fusing novel traits with deep learning features.
• To improve the accuracy of identifying individuals using only side-face information.

Main Methods:

• Proposed novel soft biometric traits based on facial anthropometry, extractable from both frontal and side face views.
• Employed a framework for fusing these soft traits with vision-based deep features (ResNet-50).
• Utilized the CMU Multi-PIE dataset for training on front faces and testing/querying with side faces.

Main Results:

• Feature-level fusion of proposed soft traits with ResNet-50 deep features significantly enhanced side face recognition performance.
• Incorporating global soft biometrics further boosted accuracy by up to 23%.
• The framework demonstrated effectiveness in zero-shot recognition scenarios.

Conclusions:

• The proposed soft biometric traits are effective for enhancing side face recognition.
• Fusion of anthropometric soft traits with deep features offers a promising approach for robust biometric identification.
• This research addresses the under-explored area of side-face recognition, improving its practical applicability.