Related Experiment Video
Updated: Aug 14, 2026
11:15
Applications of EEG Neuroimaging Data: Event-related Potentials, Spectral Power, and Multiscale Entropy
Published on: June 27, 2013
Interpretable Spectral Features for Cross-Session EEG Biometric Identification and Verification
Cai Chen1,2,3,4, Jiazheng Sun1, Danyang Lv1
1Shandong Institute of Advanced Technology, Chinese Academy of Sciences, Jinan 250100, China.
Sensors (Basel, Switzerland)
|August 13, 2026
Summary
This study introduces a new framework for electroencephalogram (EEG) biometric sensing that overcomes cross-session variability. Target-session calibration significantly enhances EEG authentication accuracy and reliability for secure human identification.
Area of Science:
- Biometrics
- Neuroscience
- Signal Processing
Background:
- Electroencephalogram (EEG)-based biometrics offer potential for secure, user-specific authentication.
- Practical EEG biometric systems face challenges due to cross-session non-stationarity and data leakage in evaluation.
- Existing methods may yield overly optimistic performance metrics due to inadequate separation of training, calibration, and testing data.
Purpose of the Study:
- To develop and evaluate a robust cross-session EEG biometric sensing framework.
- To address limitations of non-stationarity and data leakage in EEG authentication.
- To improve the accuracy and reliability of EEG-based human identification and verification.
Main Methods:
- Developed a framework using interpretable spectral features (power spectral density, differential entropy, log-variance) and target-session calibration.
- Employed strict trial-level, non-overlapping data partitioning for training, calibration, and blind testing to prevent leakage.
- Evaluated performance using various machine learning models (SVM, LDA, Logistic Regression, Random Forest, EEGNet) and metrics (Rank-N accuracy, EER, AUC, TAR).
Main Results:
- Limited target-session calibration dramatically improved Rank-1 accuracy, e.g., from 44.82% to 98.70% for SVM and 19.93% to 99.94% for LDA on an in-house dataset.
- Support Vector Machine with differential entropy achieved excellent performance: Equal Error Rate (EER) of 1.04 ± 0.60%, AUC of 0.9985 ± 0.0008, and True Acceptance Rate (TAR) at 0.1% False Acceptance Rate (FAR) of 98.44 ± 1.06%.
- External validation on a public dataset confirmed the benefits of the calibration-assisted approach.
Conclusions:
- The proposed cross-session EEG biometric framework with target-session calibration significantly enhances authentication performance.
- Strict data partitioning effectively prevents leakage, leading to more reliable evaluation of EEG biometrics.
- This approach demonstrates strong potential for practical, secure, and user-specific human identification using EEG signals.
