Abstract
Spoofing attacks remain one of the most inherent problems with traditional biometrics. Therefore, the investigation into other solutions for subject recognition is warranted. Physiological signals are recently employed as new biometrics traits that are not readily visible, like Electroencephalography (EEG), Electrocardiography (ECG), and Photoplethysmography (PPG). In particular, we are interested in PPG since its simple acquisition and its liveness clue. In this study, we have proposed an effective ppg-based biometric model compromising between minimizing the complexity of the model as regards trainable parameters while keeping high performance by using a long short-term memory (LSTM) network for the classification of ppg waveforms by modeling time series sequences. The proposed model relies on three sequential LSTM layers to capture the sequential feature of ppg recordings. Our proposed model outperforms previous state-of-the-art studies.