A New Score Level Fusion Approach for Stable User Verification System Using the PPG Signal

The recent advances in AI have made significant progress in several applications including biometric recognition. In this work, we utilize a specific biometric modality, photoplethysmography signal, for user verification systems. This physiological signal consists of user-specific features that make it suitable to authenticate a user. Yet, to be applied in realistic scenarios, time-stable features should be developed as well. Therefore, we propose a variation-stable approach tested on four score fusion techniques to find unique and time-stable features. We evaluate the proposed system on databases collected from single- and two-sessions. In the earlier, the training and testing are done solely on one session data to find user-specific features, while the second scenario is performed on data from two different sessions to investigate the time permanence of the features. The outcomes demonstrate the superiority of the proposed verification system simulated on three public datasets and one database collected for this work.