BROWSE
Detail
Journal Article
Measurement and Evaluation of Evoked Dual-Channel Ear-EEG Biometrics Leveraging Multistimuli–Response Feature Integration for Wearable Applications
Cited 0 time in 
Share 
Share
- Authors
- Taewook Kang, Mariem Kallel, Jingyo Lim, Jae-Jin Lee, Seong-Eun Kim
- Issue Date
- 2026-03
- Citation
- IEEE Transactions on Instrumentation and Measurement, v.75, pp.1-17
- ISSN
- 0018-9456
- Publisher
- IEEE
- Language
- English
- Type
- Journal Article
- Abstract
- Noninvasive electroencephalography (EEG) has emerged as a promising biometric modality, offering inherent physiological uniqueness and robustness against spoofing. However, most EEG-based biometric systems rely on high-density electrode arrays, which hinder practical deployment due to device complexity and user discomfort. Accordingly, this study investigates the effectiveness of a two-channel ear-EEG system, which acquires brain signals via in-ear sensors, for biometric identification. A compact and dedicated measurement setup was developed to acquire EEG responses to eight distinct stimuli, including steady-state visual evoked potentials (SSVEPs) and auditory signals, under a controlled laboratory environment. Using this measurement configuration, a structured and consistent dataset was obtained from 58 participants to evaluate the discriminative potential of ear-EEG signals. To identify the most effective stimuli for person identification, classification performance was analyzed across all individual stimuli, as well as with strategically designed combinations that integrated EEG responses from complementary stimulus conditions, thereby capturing synergistic patterns for enhanced discriminability. Furthermore, this study proposes a novel method for constructing an integrated EEG biometric (IEB) feature by leveraging temporal, spatial, and cross-stimulus–response integration, based on an analysis of metric effectiveness across sessions, channels, and stimulus types. Through performance evaluations of various neural network (NN) architectures optimized via Bayesian parameter tuning, a model architecture was identified that effectively leverages IEB features, achieving an average classification accuracy of 98.22% across 58 participants under fivefold CV. To enhance practical applicability, a lightweight integrated feature was constructed by combining responses from the two highly discriminative stimuli. This compact feature maintained robust performance, achieving up to 95.92% accuracy and demonstrating the efficiency of the integration strategy in balancing identification accuracy and temporal efficiency.
- Keyword
- Biometrics, electroencephalography (EEG) measurement, feature extraction, machine learning, neural networks (NNs)
- KSP Keywords
- Biometric identification, Brain signals, Classification Performance, EEG Biometrics, EEG signals, Enhanced discriminability, Feature extractioN, Feature integration, High-density, Identification accuracy, Integration strategy