Continuous Authentication with Eye Movement Biometrics

Most computer systems check user identity only at the initial login session. Continuous authentication uses data collected while using the system to check for user's identity for all session's length. This paper describes a method for authenticating user continuously using their eye movement biometrics. The new dataset was created based on data from laptop webcam, 40 features were extracted for each user. We used Siamese network to calculate the similarities of eye movement biometrics between 2 users for the task of continuous authentication. We achieved accuracy of 79.68% on video viewing task, 74.25% on slideshow task and 69.37% on reading task. Using a webcam as opposed to eye trackers allows to apply non-stop authentication without having any unique devices. Currently webcam-based eyetrackers are primarily used for behavioral analytics. However, existing Information Security studies only use this type of data for initial login (e.g., password or PIN input). Even though the achieved accuracy has some room for improvement, the suggested approach can be used as a secondary factor to help with protection against spoofing attacks.