Gazenum: unlock your phone with gaze tracking viewing numbers for authentication

Smartphones hold vast amounts of private information, making effective authentication schemes crucial to prevent privacy leaks. Knowledge-based authentication methods, such as PINs or password entry, are susceptible to side-channel attacks. While various gaze-based authentication schemes have been proposed, most require additional devices like eye trackers. Therefore, we propose GazeNum, an authentication system leveraging a smartphone’s front camera for gaze tracking. It utilizes depth-separable convolution to extract multi-scale eye image features, combined with eye contour vectors, to predict gaze points. Additionally, GazeNum employs support vector regression for personalized calibration, enhancing system accuracy. Based on the gaze point prediction model, GazeNum allows users to authenticate themselves by observing a preset password only through the front-facing camera, and the system grants access only if the gaze point and the password are correctly matched. This approach eliminates the need for additional hardware and introduces a novel, more secure authentication method through personalized calibration and multi-scale feature extraction. In a user study with 20 participants, GazeNum demonstrated 93.2% authentication accuracy and an average accuracy of 98.6% for single digits, validating its potential as an authentication system.