User Recognition System based on PPG Signal

Abstract

People are connected to various online and off-line systems that facilitate their interactions performed on a daily basis. In all these activities, individuals use their credentials to create, share and pay which could be related to the individual’s information. This sensitive data should be highly secured and thus, user recognition systems are highly empowered and invested recently.Biometric system is a technology which takes an individual’s traits as input and identifies the individual as a genuine or malicious user. In this work, our goal is to develop the verification system based on a physiological signal which has unique properties for each individual and simultaneously ensures the liveness of user since it is one of its inherent characteristics. Among diverse physiological signals, we focus on the photoplethysmography (PPG) since it can be acquired from low cost, more accessible and portable devices. The advantages correlated to the PPG signal bring it more practical and appealing to be utilized in real applications. However, the PPG signal has theoretical challenging aspects to be considered which include time variability and inherent randomness. To this end, we focus on investigating the feasibility of employing the PPG signal for person verification by developing time-stable and distinguishable features. To achieve our goal, there are five main objectives to be considered. First, a large and reliable PPG database with multi-sessions should be recorded to investigate the time stability of verification system. Second, the appropriate pipeline of PPG verification system needs to be suggested. Third, the conventional machine learning and deep learning models are investigated to understand the suitable model for PPG verification. Fourth, we develop the Generative Adversarial Network which generates the synthetic PPG with holding the unique and time-stable features. Last, we proposed the score-level fusion method of single modality (PPG) with different representations to achieve the better results with less complexity. In the collected database, we achieved up to 11.5% Equal Error Rate (EER) whereas recent works in the literature showed 23.2%, 19.1% EER in their databases. The obtained results guarantee that our PPG verification system has a high possibility to be considered in real world.