Towards Improved Eye Movement Biometrics: Investigating New Features with Neural Networks.

Area of Science:

• Biometrics and Human-Computer Interaction
• Artificial Intelligence and Machine Learning

Background:

• Secure access to resources is increasingly vital.
• Eye movement analysis offers a promising biometric modality.
• Previous research highlights the potential of eye tracking for identification.

Purpose of the Study:

• To develop and evaluate novel biometric identification methods using eye movement dynamics.
• To explore the efficacy of neural networks in analyzing eye movement patterns for user authentication.
• To assess the performance of two distinct feature extraction and classification approaches.

Main Methods:

• Two methods utilizing neural networks were developed for eye movement-based identification.
• Method 1: Feature vector from a 100-element time series of eye movement dynamics (velocity, acceleration, jerk, etc.) using Long Short-Term Memory (LSTM) networks.
• Method 2: Statistical values derived from the same eye movement dynamics, processed by dense networks.

Main Results:

• The LSTM-based method achieved a high accuracy of 96% using time series features.
• The second method, employing statistical values and dense networks, yielded 76% accuracy.
• Results were validated across a three-year span of eye movement recordings from the GazeBase dataset.

Conclusions:

• Eye movement dynamics, analyzed via neural networks, present a viable and accurate biometric solution.
• The LSTM approach demonstrates superior performance for user identification based on eye movement patterns.
• Further research is warranted to refine these methods for robust and secure access control.