NEUROAUTH-GNN: EEG-BASED EYE-BLINK CAPTCHA AUTHENTICATION USING ATTENTION-GUIDED GRAPH NEURAL NETWORKS AND ADAPTIVE NEURO-SVM

Abstract

Brain-Computer Interface (BCI) based authentication techniques have become a promising option for various secure healthcare devices, smart assistive technologies, cognitive security, human-machine interaction, etc. Because of their adaptive and user-specific authentication capabilities. Among them, EEG-based eye-blink authentication achieves hands-free and effortless user verification. For physically disabled persons, EEG-based eye-blink authentication is more helpful. Conventional EEG authentication algorithms mostly rely on removing noise and artifact, weak spatial-temporal features extraction and poor classification accuracy when there are nonlinear EEG signal distributions. Traditional machine learning classifiers are difficult to differentiate overlapping blink patterns due to highly dynamic and complex EEG signals. In this paper, we develop an EEG-based eye-blink CAPTCHA authentication scheme, namely NeuroAuth-GNN, using Attention-Guided Graph Neural Networks (GNN) and Adaptive Neuro-SVM (AN-SVM). This scheme acquires EEG signals during directional eye blinks (e.g., left, right, up, down) by a non-invasive BCI headset. After bandpass filtering and normalization on the collected EEG signals, a discriminative spatial-temporal EEG representation is extracted using an attention-guided GNN, which models the relationship between electrode connectivity and neural activity. Extracted features are classified by AN-SVM with adaptive margin and weights. Experimental results achieved better performance with higher authentication accuracy, robustness and attack resilience in contrast to prior methods and are suitable for next-generation cognitive authentication applications.