ECG Biometric Spoofing Using Adversarial Machine Learning

As the Covid-19 pandemic becomes a nationwide problem, physical contact is no longer acceptable. Therefore, biometric technology can be used for practicing social distancing to prevent the spread of the virus. However, face and fingerprint are vulnerable to presentation attacks. Hence alternative modalities such as ECG based biometric become popular. In this paper, we develop a novel presentation attack using a GAN where a short template of the victim's ECG is captured by an attacker and used to generate synthetic fake ECG signals. We also propose a novel framework utilizing residual neural network architecture to analyze ECG presentation attacks.