Brain Signal–Driven User Authentication Systems


Research Overview

Electroencephalography (EEG) offers a unique window into human cognition, capturing rapid neural responses that are difficult to observe or replicate externally. These properties make EEG an intriguing candidate for next-generation user authentication systems, particularly as traditional methods (such as passwords, fingerprints, and facial recognition) remain vulnerable to theft, spoofing, and deepfake manipulation.

Our work focuses on developing robust EEG-based user authentication pipelines using:

We aim to design authentication systems that remain reliable across days and sessions, operate seamlessly in VR/AR, and maintain user privacy without revealing personal behavioral traits.

Active Projects

Cognitive & Memory-Driven EEG-Based User Authentication

This project introduces a novel EEG-based biometric authentication system that uses cognitive and memory-related stimuli to evoke highly distinctive brainwave responses.

Key Idea: Participants performed two tasks designed to elicit strong, user-specific event-related potentials (ERPs):
  • Color Congruency Task (Stroop Task): Users viewed color words printed in matching or mismatching ink colors, producing attention-driven ERPs such as the P300.
  • Memory Recognition Task: Users memorized a set of words, then identified repeated vs. new words. This task elicited robust N400 and late positive components, producing the most distinctive neural signatures.

The study collected EEG from 34 participants across two sessions to evaluate long-term stability, using cognitive and memory-based stimuli to evoke distinctive ERPs. A rich set of spectral, temporal, spatial, and inter-hemispheric features was extracted and classified using multiple machine-learning models trained on Session 1 and tested on Session 2. Memory-driven stimuli achieved the best performance, with Logistic Regression reaching 84.07% accuracy and several users achieving perfect authentication, demonstrating strong cross-session reliability. The dataset generated from this study is available upon request for research and academic purposes.

ERP-Based User Authentication Using Ensemble Learning

This study evaluates an EEG-based user authentication system that leverages P300 and N400 event-related potentials to distinguish genuine users from impostors. Using the publicly available Brainwave Authentication Dataset, the system applies a full preprocessing pipeline (bandpass filtering, ICA artifact removal, baseline correction, and 1-second segmentation) followed by comprehensive feature extraction, including PSD features across five frequency bands and DWT-based statistical features. Multiple classifiers were examined, with ensemble models showing superior performance. Across 38 participants and five stimulus categories, the best result achieved an average EER of 2.53% using CatBoost on the N400-Words task, significantly outperforming prior studies and demonstrating the strength of semantic-processing ERPs for authentication.

EEG-Based User Authentication Using Face Familiarity in VR

This ongoing project explores whether neural responses to familiar and unfamiliar faces inside virtual reality can serve as a secure and unobtrusive form of user authentication. When people see familiar faces, their brains automatically generate characteristic event-related potentials—such as N170, N250, and N400—that are difficult to fake or consciously control, making them strong candidates for implicit biometric signals. We are developing a VR-integrated EEG acquisition framework that synchronizes face presentations with mobile EEG systems and extracts cognitive and perceptual features associated with face processing. The project focuses on creating an authentication pipeline tailored for immersive environments, examining the feasibility, stability, and privacy implications of using face familiarity as an authentication cue. Data collection and analysis are currently in progress, and the resulting dataset will be made available upon request once the study is completed.

Resources