Understanding Gaze-Based Identification in VR Through Preattentive Processing and Binocular Rivalry

Paper Title

Understanding Gaze-Based Identification in VR Through Preattentive Processing and Binocular Rivalry

Publication Info

• Topic area: Gaze-based biometrics for user identification and authentication in VR.
• Keywords: Gaze-based biometrics, virtual reality, preattentive processing, binocular rivalry, user identification, authentication, usability, VR-native stimuli, eye tracking, biometric security.

Background and Problem

• Problem / challenge: Existing gaze-based biometrics in VR rely on generic stimuli and do not fully leverage VR-specific features like binocular rivalry and stereoscopic displays. Temporal robustness and usability in real-world contexts remain underexplored.
• Significance: Gaze-based biometrics offer hands-free, secure, and intuitive authentication in VR, addressing usability and security challenges of traditional methods like PINs or out-of-band authentication.
• Motivation and related work: Prior work has demonstrated the feasibility of gaze biometrics but has focused on non-immersive or generic stimuli. Few studies have explored VR-native stimuli or systematically evaluated usability and performance across multiple days with commercial devices.

Solution

• Proposed approach: Development and evaluation of VR-specific visual stimuli leveraging preattentive processing and binocular rivalry to elicit distinctive gaze responses for biometric identification.
• Novelty:
  1. Introduction of VR-native stimuli (NonBRiv and BRiv) designed to exploit preattentive processing and binocular rivalry.
  2. Multi-day evaluation of biometric discriminability using commercial VR headsets (Meta Quest Pro).
  3. Usability assessment of gaze-based login methods compared to traditional authentication techniques.
  4. Exploration of user perceptions, transparency, and fallback mechanisms in gaze-based authentication.
• Procedure and key techniques:
  1. Design of NonBRiv (non-rivalrous) and BRiv (binocular rivalry) stimuli with controlled visual features.
  2. Multi-day data collection from 26 participants, analyzing gaze dynamics under both conditions.
  3. Evaluation of identification accuracy using classifiers (SVM, LSTM, EKYT) and cross-validation.
  4. Usability study with 16 participants comparing gaze-based methods to PIN and OOBA in a Wizard-of-Oz login scenario.

Results

• Concrete findings:
  • EKYT classifier achieved 50.7% accuracy with BRiv and 70.0% accuracy with Hybrid (aggregated samples).
  • BRiv stimuli elicited higher gaze velocities and richer discriminatory cues than NonBRiv.
  • Single-day enrollment yielded low error rates (EER=3.6% for NonBRiv), but performance degraded over multiple days (EER=25-33%).
  • Gaze-based methods were faster (8-9 seconds) than PIN (11.25 seconds) and OOBA (17.35 seconds) for login completion.
• Advantage over baselines:
  • Gaze-based methods outperformed PIN and OOBA in hedonic quality (HQ) and overall user experience (UEQ-S).
  • Hybrid stimuli reduced cross-day variability compared to NonBRiv and BRiv alone.
• Experiments / evaluation:
  • User study I: Multi-day biometric evaluation with 26 participants, collecting 44,928 trials.
  • User study II: Usability and perception study with 16 participants, comparing gaze-based methods to PIN and OOBA.
  • Metrics: Identification accuracy, equal error rate (EER), UEQ-S scores, cybersickness, and completion time.
• Limitations and future work:
  • Performance degradation across days and under real-world conditions.
  • Limited sample size (26 participants) and controlled lab environment.
  • Need for online calibration, replay attack defenses, and personalized stimuli.
  • Exploration of binocular rivalry dynamics and their contribution to discriminability.

Summary

This study introduces VR-native gaze-based biometrics leveraging preattentive processing and binocular rivalry, achieving meaningful identification accuracy (up to 70.0% with Hybrid stimuli) and demonstrating usability advantages over traditional methods. While gaze-based methods were perceived as fast and intuitive, challenges such as cross-day robustness, visual comfort, and transparency remain. The findings provide design guidelines for improving gaze-based authentication in VR, emphasizing the need for adaptive calibration, explainable feedback, and fallback mechanisms. Future work should address real-world deployment challenges and explore richer gaze features for enhanced performance.