ETD

• ppg
• rppg

This thesis investigates biometric identity verification based on photoplethysmography (PPG) and remote photoplethysmography (rPPG) signals.
Finger-PPG signals from 80 subjects, drawn from a publicly available dataset, were used to train a one-versus-all identity verification framework based on a neural network model. For each subject, two physiological conditions were considered—resting state and post-exercise—to evaluate the robustness of the biometric system to exercise-induced physiological variability.
Experimental results indicate that contact-based PPG provides more reliable biometric performance. Concerning the resting state condition, the system achieved an average false acceptance rate (FAR) of 5% and a false rejection rate (FRR) of 47%, while in the post-exercise phase the FAR increased to 24%, highlighting challenges in generalization under unseen conditions. Remote PPG signals were extracted from a subset of RGB facial videos of 50 subjects and analyzed using the same verification framework. Compared to contact-based PPG, rPPG exhibited substantially lower usability, with an average FAR of 5% and an FRR of 78% during rest, reflecting limitations related to signal quality and physiological noise. Performance further deteriorated when testing the rPPG system under post-exercise conditions, indicating that the current rPPG-based approach is not sufficiently reliable for practical biometric identity verification.