PPG-based Personalized Verification System

Dae Yon Hwang, Dimitrios Hatzinakos

Abstract: Convergence between online and off-line systems gives us a great chance to enrich our societies but it also requires a high secure system to verify true user from fraud. In this paper, we propose a novel deep learning-based verification model using Photoplethysmography (PPG) signals. The goal of this paper is to build a personalized data-driven network by employing convolution neural network (CNN) with long-short term memory (LSTM), to model the time-series sequence inherent within the PPG signal. After building each personalized network, each network can be applied to distinguish a true user from others. The proposed network was evaluated on the BioSec. Lab PPG dataset at University of Toronto, which achieved an average of 10-fold cross-validation accuracy of 96% (in single-session) and 72.7% (in two-sessions).