Real-Time PPG-Based Biometric Identification: Advancing Security with 2D Gram Matrices and Deep Learning Models.

Area of Science:

• Biometrics and Security Engineering
• Signal Processing and Machine Learning

Background:

• Liveness detection is vital for biometric system security against spoofing.
• Traditional methods face challenges, necessitating advanced, spoof-resistant alternatives.
• Photoplethysmography (PPG) signals present a promising, non-invasive biometric modality.

Purpose of the Study:

• To evaluate the effectiveness of PPG signals for liveness detection in biometric authentication.
• To develop and validate a robust system for real-time biometric identification using PPG.
• To enhance security by leveraging the inherent spoof-resistance of PPG signals.

Main Methods:

• Collected PPG signals from 40 subjects using a custom acquisition system.
• Transformed PPG signals into 2D representations via Gram matrix conversion.
• Employed an EfficientNetV2 B0 model with an LSTM network for analysis and authentication.

Main Results:

• Achieved 99% accuracy on the test set for biometric authentication.
• Demonstrated high precision, recall, and F1 scores, indicating robust performance.
• Successfully validated the model in real-time identification scenarios.

Conclusions:

• PPG signals are a cost-efficient and highly spoof-resistant biometric source.
• The developed EfficientNetV2 B0-LSTM model offers a superior solution for next-generation biometric systems.
• The system provides enhanced security and effectiveness for biometric recognition.