Photoplethysmogram Biometric Authentication Using a 1D Siamese Network.

Area of Science:

• Biometrics and Human-Computer Interaction
• Wearable Technology and Signal Processing

Background:

• Conventional input devices are unsuitable for metaverse and virtual reality (VR) head-mounted display environments.
• There is a need for nonintrusive, continuous biometric authentication for immersive digital experiences.

Purpose of the Study:

• To propose a novel biometric identification model using photoplethysmogram (PPG) signals from wrist wearable devices.
• To evaluate the effectiveness of a multicycle averaging method for noise reduction and characteristic preservation in PPG signals.
• To validate the performance of a one-dimensional Siamese network for biometric authentication.

Main Methods:

• A one-dimensional Siamese network model was developed for biometric identification.
• A multicycle averaging technique was employed for PPG signal preprocessing, avoiding traditional filters.
• The number of averaging cycles was varied to optimize noise reduction and data fidelity.
• Genuine and impostor PPG data were utilized for performance evaluation.

Main Results:

• The multicycle averaging method effectively reduced noise while preserving unique personal characteristics.
• The one-dimensional Siamese network demonstrated strong performance in distinguishing between genuine and impostor subjects.
• The optimal configuration involved five overlapping cycles, yielding an AUC score of 0.988 and accuracy of 0.9723.
• The proposed method showed high accuracy irrespective of the number of enrolled subjects.

Conclusions:

• The proposed PPG-based biometric identification model offers time-efficient and secure authentication for wearable devices.
• Multicycle averaging is a viable alternative to traditional filtering for PPG signal enhancement.
• The one-dimensional Siamese network provides a robust framework for continuous, nonintrusive biometric authentication in VR/metaverse applications.