Slide
Centre Interdisciplinaire
de Recherche et d’Innovation
en Cybersécurité et Société
de Recherche et d’Innovation
en Cybersécurité et Société
1.
Dang, L. B.; Truong, V. Tuan; Le, L. B.; Falk, T. H.
EM-Auth: A Generalizable Eye-Movement Authentication Framework using Event Cameras Proceedings Article
In: IEEE Conf. Artif. Intell., CAI, pp. 2064–2069, Institute of Electrical and Electronics Engineers Inc., 2026, ISBN: 979-833156039-3 (ISBN), (Journal Abbreviation: IEEE Conf. Artif. Intell., CAI).
Abstract | Links | BibTeX | Tags: augmented reality, Authentication, Cameras, Convolutional neural networks, Data privacy, Eye movement patterns, Eye movements, eye tracking, Eye-tracking, Frame-based, High power consumption, Neuromorphic vision sensors, Power potential, Privacy concerns, Temporal resolution, Transfer learning, User authentication, Virtual and augmented reality, virtual reality
@inproceedings{dangEMAuthGeneralizableEyeMovement2026,
title = {EM-Auth: A Generalizable Eye-Movement Authentication Framework using Event Cameras},
author = {L. B. Dang and V. Tuan Truong and L. B. Le and T. H. Falk},
url = {https://www.scopus.com/pages/publications/105042114237?origin=resultslist},
doi = {10.1109/CAI68641.2026.11536355},
isbn = {979-833156039-3 (ISBN)},
year = {2026},
date = {2026-01-01},
booktitle = {IEEE Conf. Artif. Intell., CAI},
pages = {2064–2069},
publisher = {Institute of Electrical and Electronics Engineers Inc.},
abstract = {With the rapid emergence of virtual and augmented reality (VR/AR) head-mounted devices that closely interact with users' eyes, eye-tracking-based applications are becoming increasingly promising. Among them, eye movement patterns have emerged as a reliable biometric modality for user authentication. However, existing solutions based on conventional frame-based cameras face significant limitations, including low temporal resolution, high latency, high power consumption, and potential privacy concerns due to the detailed appearance captured around the eyes. To address these challenges, we propose EM-Auth, a novel framework for eye-tracking-based user authentication that leverages event cameras, a type of neuromorphic vision sensor providing asynchronous, high-temporal-resolution data, while preserving user privacy. Our method introduces a two-stage spatiotemporal feature extraction pipeline: first, a lightweight convolutional neural network (CNN) extracts spatial features from event streams; second, a transformer-based module models the temporal dependencies to generate robust eye representations. Extensive experiments show that our framework significantly outperforms existing state-of-the-art (SOTA) methods, achieving 0.008 and 0.015 equal error rates (EERs) on two representative datasets. Notably, EM-Auth demonstrates a strong generalization capability. In a zero-shot setting where the model is trained on a specific dataset and evaluated on previously unseen users, our approach achieves authentication accuracies of 98.4% and 95.9% on the EV-Eye and EBV-Eye datasets, respectively. Under cross-dataset evaluation, our framework achieves an authentication accuracy of nearly 85% without additional fine-tuning, and consistently improves performance through transfer learning. © 2026 IEEE.},
note = {Journal Abbreviation: IEEE Conf. Artif. Intell., CAI},
keywords = {augmented reality, Authentication, Cameras, Convolutional neural networks, Data privacy, Eye movement patterns, Eye movements, eye tracking, Eye-tracking, Frame-based, High power consumption, Neuromorphic vision sensors, Power potential, Privacy concerns, Temporal resolution, Transfer learning, User authentication, Virtual and augmented reality, virtual reality},
pubstate = {published},
tppubtype = {inproceedings}
}
With the rapid emergence of virtual and augmented reality (VR/AR) head-mounted devices that closely interact with users' eyes, eye-tracking-based applications are becoming increasingly promising. Among them, eye movement patterns have emerged as a reliable biometric modality for user authentication. However, existing solutions based on conventional frame-based cameras face significant limitations, including low temporal resolution, high latency, high power consumption, and potential privacy concerns due to the detailed appearance captured around the eyes. To address these challenges, we propose EM-Auth, a novel framework for eye-tracking-based user authentication that leverages event cameras, a type of neuromorphic vision sensor providing asynchronous, high-temporal-resolution data, while preserving user privacy. Our method introduces a two-stage spatiotemporal feature extraction pipeline: first, a lightweight convolutional neural network (CNN) extracts spatial features from event streams; second, a transformer-based module models the temporal dependencies to generate robust eye representations. Extensive experiments show that our framework significantly outperforms existing state-of-the-art (SOTA) methods, achieving 0.008 and 0.015 equal error rates (EERs) on two representative datasets. Notably, EM-Auth demonstrates a strong generalization capability. In a zero-shot setting where the model is trained on a specific dataset and evaluated on previously unseen users, our approach achieves authentication accuracies of 98.4% and 95.9% on the EV-Eye and EBV-Eye datasets, respectively. Under cross-dataset evaluation, our framework achieves an authentication accuracy of nearly 85% without additional fine-tuning, and consistently improves performance through transfer learning. © 2026 IEEE.
2.
Conti, J.; Ozell, B.; Paquette, E.; Renaud, P.
Adjusting stereoscopic parameters by evaluating the point of regard in a virtual environment Journal Article
In: Computers and Graphics (Pergamon), vol. 69, pp. 24–35, 2017, ISSN: 00978493 (ISSN), (Publisher: Elsevier Ltd).
Abstract | Links | BibTeX | Tags: Cameras, Cave automatic virtual environments, Cell proliferation, Depth of field blur, Depth-of-field blur, eye tracking, Immersive virtual environments, Inter-camera distance, Point of regard tracking, Point of regards, Research and development, Stereo image processing, Stereoscopic parameters, Subjective evaluations, virtual reality
@article{conti_adjusting_2017,
title = {Adjusting stereoscopic parameters by evaluating the point of regard in a virtual environment},
author = {J. Conti and B. Ozell and E. Paquette and P. Renaud},
url = {https://www.scopus.com/inward/record.uri?eid=2-s2.0-85032854314&doi=10.1016%2fj.cag.2017.08.017&partnerID=40&md5=529b21bdd959c209d79b5262fae53516},
doi = {10.1016/j.cag.2017.08.017},
issn = {00978493 (ISSN)},
year = {2017},
date = {2017-01-01},
journal = {Computers and Graphics (Pergamon)},
volume = {69},
pages = {24–35},
publisher = {Elsevier Ltd},
abstract = {Despite the growth in research and development in the area of virtual reality over the past few years, virtual worlds do not yet convey a feeling of presence that matches reality. This is particularly due to the difference in visual perception of flat images as compared to actual 3D. We studied the impact of two parameters of the stereoscopic configuration, namely, the inter-camera distance (ICD) and the presence of a depth of field blur (DOF blur). We conducted an experiment involving 18 participants in order to evaluate this impact, based on both subjective and objective criteria. We examined six configurations which differed in the presence or absence of DOF blur and the value of the ICD: fixed and equal to the anatomical interpupillary distance, fixed and chosen by the participant, or variable, depending on the depth of the viewer's point of regard (POR). The DOF blur and the variable ICD require the use of an eye tracking system in order to be adjusted with respect to the POR. To our knowledge, no previously published research has tested a gaze-contingent variable ICD along with dynamic DOF blur in a Cave Automatic Virtual Environment. Our results show that the anatomical and variable ICD performed similarly regarding each criterion of the experiment, both being more efficient than the fixed ICD. Besides, as with earlier similar attempts, the configurations with DOF blur obtained lower subjective evaluations. Although mainly not significant, the results obtained by the variable ICD and DOF blur are likely due to a noticeable delay in the parameters update. We also designed a new methodology to objectively compare the geometry and depth rendering, based on the reproduction of the same scene in the real and virtual setups, and then on the study of resulting ocular convergence and angular deviation from a target. This leads to a new comparative criterion for the perceptual realism of immersive virtual environments based on the visual behavior similarity between real and virtual setups. © 2017 Elsevier Ltd},
note = {Publisher: Elsevier Ltd},
keywords = {Cameras, Cave automatic virtual environments, Cell proliferation, Depth of field blur, Depth-of-field blur, eye tracking, Immersive virtual environments, Inter-camera distance, Point of regard tracking, Point of regards, Research and development, Stereo image processing, Stereoscopic parameters, Subjective evaluations, virtual reality},
pubstate = {published},
tppubtype = {article}
}
Despite the growth in research and development in the area of virtual reality over the past few years, virtual worlds do not yet convey a feeling of presence that matches reality. This is particularly due to the difference in visual perception of flat images as compared to actual 3D. We studied the impact of two parameters of the stereoscopic configuration, namely, the inter-camera distance (ICD) and the presence of a depth of field blur (DOF blur). We conducted an experiment involving 18 participants in order to evaluate this impact, based on both subjective and objective criteria. We examined six configurations which differed in the presence or absence of DOF blur and the value of the ICD: fixed and equal to the anatomical interpupillary distance, fixed and chosen by the participant, or variable, depending on the depth of the viewer's point of regard (POR). The DOF blur and the variable ICD require the use of an eye tracking system in order to be adjusted with respect to the POR. To our knowledge, no previously published research has tested a gaze-contingent variable ICD along with dynamic DOF blur in a Cave Automatic Virtual Environment. Our results show that the anatomical and variable ICD performed similarly regarding each criterion of the experiment, both being more efficient than the fixed ICD. Besides, as with earlier similar attempts, the configurations with DOF blur obtained lower subjective evaluations. Although mainly not significant, the results obtained by the variable ICD and DOF blur are likely due to a noticeable delay in the parameters update. We also designed a new methodology to objectively compare the geometry and depth rendering, based on the reproduction of the same scene in the real and virtual setups, and then on the study of resulting ocular convergence and angular deviation from a target. This leads to a new comparative criterion for the perceptual realism of immersive virtual environments based on the visual behavior similarity between real and virtual setups. © 2017 Elsevier Ltd
3.
Conti, J.; Ozell, B.; Paquette, E.; Renaud, P.
Adjusting stereoscopic parameters by evaluating the point of regard in a virtual environment Journal Article
In: Computers and Graphics (Pergamon), vol. 69, pp. 24–35, 2017, ISSN: 00978493 (ISSN).
Abstract | Links | BibTeX | Tags: Cameras, Cave automatic virtual environments, Cell proliferation, Depth of field blur, Depth-of-field blur, eye tracking, Immersive virtual environments, Inter-camera distance, Point of regard tracking, Point of regards, Research and development, Stereo image processing, Stereoscopic parameters, Subjective evaluations, virtual reality
@article{contiAdjustingStereoscopicParameters2017,
title = {Adjusting stereoscopic parameters by evaluating the point of regard in a virtual environment},
author = {J. Conti and B. Ozell and E. Paquette and P. Renaud},
url = {https://www.scopus.com/inward/record.uri?eid=2-s2.0-85032854314&doi=10.1016%2fj.cag.2017.08.017&partnerID=40&md5=529b21bdd959c209d79b5262fae53516},
doi = {10.1016/j.cag.2017.08.017},
issn = {00978493 (ISSN)},
year = {2017},
date = {2017-01-01},
journal = {Computers and Graphics (Pergamon)},
volume = {69},
pages = {24–35},
publisher = {Elsevier Ltd},
abstract = {Despite the growth in research and development in the area of virtual reality over the past few years, virtual worlds do not yet convey a feeling of presence that matches reality. This is particularly due to the difference in visual perception of flat images as compared to actual 3D. We studied the impact of two parameters of the stereoscopic configuration, namely, the inter-camera distance (ICD) and the presence of a depth of field blur (DOF blur). We conducted an experiment involving 18 participants in order to evaluate this impact, based on both subjective and objective criteria. We examined six configurations which differed in the presence or absence of DOF blur and the value of the ICD: fixed and equal to the anatomical interpupillary distance, fixed and chosen by the participant, or variable, depending on the depth of the viewer's point of regard (POR). The DOF blur and the variable ICD require the use of an eye tracking system in order to be adjusted with respect to the POR. To our knowledge, no previously published research has tested a gaze-contingent variable ICD along with dynamic DOF blur in a Cave Automatic Virtual Environment. Our results show that the anatomical and variable ICD performed similarly regarding each criterion of the experiment, both being more efficient than the fixed ICD. Besides, as with earlier similar attempts, the configurations with DOF blur obtained lower subjective evaluations. Although mainly not significant, the results obtained by the variable ICD and DOF blur are likely due to a noticeable delay in the parameters update. We also designed a new methodology to objectively compare the geometry and depth rendering, based on the reproduction of the same scene in the real and virtual setups, and then on the study of resulting ocular convergence and angular deviation from a target. This leads to a new comparative criterion for the perceptual realism of immersive virtual environments based on the visual behavior similarity between real and virtual setups. © 2017 Elsevier Ltd},
keywords = {Cameras, Cave automatic virtual environments, Cell proliferation, Depth of field blur, Depth-of-field blur, eye tracking, Immersive virtual environments, Inter-camera distance, Point of regard tracking, Point of regards, Research and development, Stereo image processing, Stereoscopic parameters, Subjective evaluations, virtual reality},
pubstate = {published},
tppubtype = {article}
}
Despite the growth in research and development in the area of virtual reality over the past few years, virtual worlds do not yet convey a feeling of presence that matches reality. This is particularly due to the difference in visual perception of flat images as compared to actual 3D. We studied the impact of two parameters of the stereoscopic configuration, namely, the inter-camera distance (ICD) and the presence of a depth of field blur (DOF blur). We conducted an experiment involving 18 participants in order to evaluate this impact, based on both subjective and objective criteria. We examined six configurations which differed in the presence or absence of DOF blur and the value of the ICD: fixed and equal to the anatomical interpupillary distance, fixed and chosen by the participant, or variable, depending on the depth of the viewer's point of regard (POR). The DOF blur and the variable ICD require the use of an eye tracking system in order to be adjusted with respect to the POR. To our knowledge, no previously published research has tested a gaze-contingent variable ICD along with dynamic DOF blur in a Cave Automatic Virtual Environment. Our results show that the anatomical and variable ICD performed similarly regarding each criterion of the experiment, both being more efficient than the fixed ICD. Besides, as with earlier similar attempts, the configurations with DOF blur obtained lower subjective evaluations. Although mainly not significant, the results obtained by the variable ICD and DOF blur are likely due to a noticeable delay in the parameters update. We also designed a new methodology to objectively compare the geometry and depth rendering, based on the reproduction of the same scene in the real and virtual setups, and then on the study of resulting ocular convergence and angular deviation from a target. This leads to a new comparative criterion for the perceptual realism of immersive virtual environments based on the visual behavior similarity between real and virtual setups. © 2017 Elsevier Ltd