Research Course:
Title: Filtering Effects on Eye Movement Biometrics for User Authentication
Presenter: Mehedi Raju
Advisor: Oleg Komogortsev
Date/Time: Thursday, December 8th at 11:00 a.m.
Location: https://txstate.zoom.us/j/91243653692
Abstract: Filters are used to remove high frequency noise components from eye movement data. For video-oculography systems, Stampe proposed two ”heuristic” filters in 1993. Top eye-tracker manufacturers have employed these filters as an option for recording eye movements. We demonstrate that digital low pass filters dramatically outperform the heuristic filters for unfiltered EyeLink 1000 signals. Based on the literature, which has employed various eye-tracking technologies, and our analysis of our EyeLink 1000 data, we conclude that the highest signal frequency content needed for most eye-tracking studies (i.e., to accurately measure saccades, microsaccades and smooth pursuit) is around 100 Hz, excluding fixation microtremor. We test two zero-phase low-pass digital filters, one with a cutoff of 50 Hz and one with a cutoff of 100 Hz. We perform a Fourier (FFT) analysis to examine the frequency content for unfiltered data, heuristic filtered data, and digitally filtered data. We also examine the frequency response of these filters. Our analysis states that the digital filter with a 100 Hz cutoff dramatically outperforms both heuristic filters because the heuristic filters leave noise above 100 Hz. We designed our current experiment to determine the biometric performance achievable with eye movement signals filtered with various low-pass digital filters based on the findings from. To achieve state-of-the-art biometric performance for userauthentication via eye movements, it is our objective to identify the highest signal frequency content required. For user authentication, we must determine the optimal cutoff frequency at which biometric performance excels. The results of this investigation will help to establish the minimum sampling rate needed for user authentication using eye movement biometrics. In our current experiment, we use the Eye Know You Too network architecture from and a publicly available dataset.
Deadline: Jan. 4, 2023, midnight