The recent COVID-19 outbreak has highlighted the importance of contactless authentication methods, such as those based on eye or gaze features. These techniques have the advantage that they can also be used by people wearing mouth and nose masks, which would make traditional face recognition approaches difficult to apply. Moreover, they can be used in addition to traditional authentication solutions, such as those based on passwords or PINs. In this work, we propose a study on gaze-based soft biometrics exploiting simple animations as visual stimuli. Specifically, we consider four animations in which small squares move according to different patterns and trajectories. No preliminary calibration of the eye tracking device is required. The collected data were analyzed using machine learning algorithms for both identification and verification tasks. The obtained results are particularly interesting in the verification case, that is the natural application of a soft biometric system, with accuracy scores always higher than 80 and Equal Error Rate (EER) values often lower than 10.
展开▼
