Investigation of t-SNE and dynamic time warping within a unified framework for resting-state and minor analysis visual task-related EEG alpha frequency in biometric authentication: A detailed analysis


Ozturk H., Eraslan B., Gorur K.

DIGITAL SIGNAL PROCESSING: A REVIEW JOURNAL, vol.160, no.105042, pp.1-30, 2025 (SCI-Expanded)

  • Publication Type: Article / Article
  • Volume: 160 Issue: 105042
  • Publication Date: 2025
  • Doi Number: 10.1016/j.dsp.2025.105042
  • Journal Name: DIGITAL SIGNAL PROCESSING: A REVIEW JOURNAL
  • Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Academic Search Premier, Aerospace Database, Applied Science & Technology Source, Communication Abstracts, Compendex, Computer & Applied Sciences, INSPEC
  • Page Numbers: pp.1-30
  • Ankara Yıldırım Beyazıt University Affiliated: Yes

Abstract

This study presents a dynamic biometric authentication system using EEG alpha frequency signals (8–13 Hz) recorded during resting states and visual attention tasks. Unlike static systems, EEG-based authentication detects aliveness through dynamic, continuously changing signal patterns. The alpha band was extracted using Discrete Wavelet Transform (DWT) for both resting-state and visual attention EEG data. Our framework integrates tDistributed Stochastic Neighbor Embedding (t-SNE) and Dynamic Time Warping (DTW) for feature extraction and recognition. t-SNE captures unique temporal patterns, while DTW measures Euclidean distances between resting-state EEG responses, distinguishing individuals with subject-specific and common thresholds to enhance reliability. For visual task-related EEG, a t-SNE-based thresholding mechanism employs both subject-specific and common-thresholds for robust decision-making. This approach was tested on six participants, demonstrating the effectiveness of combining thresholding logic with t-SNE feature extraction. Minor Analysis: Visual task-related EEG biometric results were included as a secondary analysis, highlighting the system’s reliability and flexibility under different task conditions. Robustness and reliability were further enhanced by integrating ensembled 1DCNN models with pattern-based, statistical-based, observation-based, and distance-based recognition methods. Testing on two frontal channels (Fp1, Fp2) from twenty participants yielded promising results: Subject 7 ach ieved a True Acceptance Rate (TAR) of 100 % and a False Acceptance Rate (FAR) of 0 %, while Subjects 15 and 2 exceeded 98 % accuracy. These findings demonstrate the system’s reliability and effectiveness in biometric authentication.