An Eye Tracking Study on Reading-Based Biometrics

Abstract In the 21st century digital information era, where data is considered the most valuable asset,  authorization and privacy concerns when accessing the data are among the biggest challenges to handle. Many methods have been developed for this purpose, and biometrics is one of them. There are several types of biometrics, such as finger prints,  face, signature, voice, and iris recognition, DNA matching, etc.
This thesis, in particular, focuses on behavioural biometrics, and specifically on reading-based biometrics through eye tracking. Using a low-price  eye tracker with a frequency of 60Hz, reading data from 31 individuals were obtained and analysed. The stimuli were 90 short sentences displayed for six seconds each, organized in three sessions of 30 sentences.  The gathered gaze data has been carefully studied using machine learning approaches, for the two typical biometric processes of identification and verification.  Different classifiers have been exploited, and their performance has been evaluated in terms of accuracy and precision. The obtained results can be considered satisfactory in the context of “soft biometrics”, which does not require extremely high accuracies as is employed together with other identification or verification techniques.