The world’s leading publication for data science, AI, and ML professionals.

The Pipeline and Limitations of the EEG-based Authentication Systems

A novel approach to integrate speech recognition into authentication systems Part 2

AI APPLICATIONS

Photo by Matthew Hume from Unsplash
Photo by Matthew Hume from Unsplash

In the previous part, we reviewed the most common authentication systems that exist. We realized the limitations of textual and graphical passwords and biometric methods, including iris, fingerprint, and facial recognition. Therefore, in this part, we will dive deeper into one of the more feasible methods in the biometric Authentication domain: EEG-based authentication systems.


1. The four steps

EEG has been widely used in industries and researches. They are involved in the entertainment of gaming interaction, robot control, emotion recognition, fatigue detection, sleep quality assessment, and clinical fields, such as abnormal brain disease detection and prediction, including Parkinson’s disease and Alzheimer’s disease [10]. Using EEG signals in authentication is firstly introduced by Thorpe et al. [11]; the authors presented a novel idea to use brainwaves for user authentication called Pass-thought. EEG-based authentication systems mainly consist of four steps: data acquisition, pre-processing, feature extraction and classification, as shown in the image below.

Four primary modules for an EEG biometric authentication system. Image from the research paper [10]
Four primary modules for an EEG biometric authentication system. Image from the research paper [10]

1.1 Data acquisition

EEG acquisition protocols can be divided into relaxation with eyes open or eyes closed, visual stimulation, mental tasks (imagination of doing specific tasks), and emotional stimuli[12].

1.2 Data preprocessing

Pre-processing stage involves different processing techniques for the signal to be ready for feature extraction. Some of the methods are: signal segmentation or framing in different epochs length, signal filtering to extract certain EEG bands, noise and artifacts, and removing artifacts. Alomari et al. [13]applied Butterworth filter with a passband of 0.1Hz – 40Hzto EEG data, then segmented, sorted and conducted artifact detection on the data to investigate the correlation between the EEG signals measured during the presentation of passwords, and how users perceived the passwords’ memorability. Their project indicates that it may be possible to predict subsequent password recall based on EEG activity during password presentation.

1.3 Feature extraction

Feature extraction is a critical part of the processing of EEG signals. The techniques have been conducted in different domains such as frequency features, time-frequency features, statistical features, entropy-based features and higher-order crossing (HOC) features. Zeynali et al. [14] proposed a modality for decreasing the error of the EEG-based key generation process by using Discrete Fourier Transform, Discrete Wavelet Transform, Auto-Regressive Modelling, Energy Entropy and Sample Entropy. Petrantonakis [15] applied HOC feature extraction analysis on EEG data to recognize emotions.

The most widely used Matlab toolbox for EEG data processing is EEGLAB, which provides an interactive graphic user interface for users to apply Independent Component Analysis(ICA), time/frequency analysis (TFA) and standard averaging methods to the recorded brain signals; Open-source Real-time EEG Source-mapping Toolbox (REST) is another useful tool for online artifact rejection and feature extraction [10].

1.4 Classification

Most current EEG-based brain-computer interfaces (BCIs)are based on Machine Learning algorithms, and the recently designed classification algorithms for EEG-based BCIs can be divided into four main categories: adaptive classifiers, matrix and tensor classifiers, transfer learning and deep learning, plus a few other miscellaneous classifiers [16]. Neural Networks(NN) and Support Vector Machines (SVM) have been used to check the correlation of password memorability and recall by Alomari et al. [17] [13]. A multinomial classification with a one-hot encoding technique was proposed to recognize a user’s brain passwords by Yang [18].

2. Limitations

2.1 The scale of the EEG database

Most EEG analyzing tasks use a small dataset. Alomari et al. [13] gained EEG data recorded using the Muse headband from 77 university volunteers via email advertisements; in their another research work [17], they gained data through conducting a lab study on 19 university students by recording data when presenting two sets of Passwords in front of the students. The sample size is particularly limited.

Some EEG databases are available publicly for human recognition purposes, such as the UCI KDD EEG database, EEG motor movement/imagery dataset, and Australian database. Still, the dataset scale cannot determine the proposed systems can be generalized to a large-scale EEG data performance. Thus a more significant population needs to be collected and tested under different data acquisition scenarios to prove that EEG signals are unique to individuals to be regarded the identification of each person.

2.2 The impact of mental health

In most experiments, brain signals of healthy people are tested as samples; however, mental disorder exists in a real-life scenario, and it may alter brain wave shapes, thus changing the results of classification models. More research needs to be done regarding the impact of mental disorders on brainwave signals and the performance of EEG authentication results.

2.3 EEG devices

Most accurate EEG devices need to put multiple electrodes on the scalp, which is inconvenient for commercial use. With technology development, more dry electrodes portable devices are invented and used in everyday life. An overview of EEG devices is listed in the table below; it specifies the sensors type(whether dry or wet), number of channels, sampling rate, weight and more features that correspond to each product. The main disadvantage of dry devices is that their accuracy is not as precise as other BCI devices used in medical areas. So whether portable devices with dry electrodes are appropriate choices for recording EEG data for identifying individuals is worth more research.

Overall of EEG devices. Image from the paper [10]
Overall of EEG devices. Image from the paper [10]

For the next part of this series, I will talk more about voice-based authentication systems and their limitations, then propose a novel authentication system integrating speech recognition.

Stay tuned and welcome to leave a comment and connect with me on Linkedin.

Fangyi Yu – Mentor – DeepLearning.AI | LinkedIn

References after Part 1:

[10] X. Gu, Z. Cao, A. Jolfaei, P. Xu, D. Wu, T.-P. Jung, and C.-T. Lin, "Eeg-based brain-computer interfaces (bcis): A survey of recent studies on signal sensing technologies and computational intelligence approaches and their applications," IEEE/ACM transactions on computational biology and bioinformatics, 2021.

[11] J. Thorpe, P. C. Van Oorschot, and A. Somayaji, "Pass-thoughts: authenticating with our minds," in Proceedings of the 2005 workshop on New security paradigms, 2005, pp. 45–56.

[12] M. Abo-Zahhad, S. M. Ahmed, and S. N. Abbas, "State-of-the-art methods and future perspectives for personal recognition based on electroencephalogram signals," IET Biometrics, vol. 4, no. 3, pp. 179–190, 2015.

[13] R. Alomari and M. V. Martin, "Classification of EEG signals using neural networks to predict password memorability," in 2018 17th IEEE International Conference on Machine Learning and Applications (ICMLA). IEEE, 2018, pp. 791–796.

[14] M. Zeynali, H. Seyedarabi, and B. Mozaffari Tazehkand, "Development of a unique biometric-based cryptographic key generation with repeatability using brain signals," Journal of AI and Data Mining, vol. 8, no. 3, pp. 343–356, 2020.

[15] P. C. Petrantonakis and L. J. Hadjileontiadis, "Emotion recognition from EEG using higher-order crossings," IEEE Transactions on information technology in Biomedicine, vol. 14, no. 2, pp. 186–197, 2009.

[16] F. Lotte, L. Bougrain, A. Cichocki, M. Clerc, M. Congedo, A. Rako-commonly, and F. Yger, "A review of classification algorithms for EEG-based brain-computer interfaces: a 10-year update," Journal of neural engineering, vol. 15, no. 3, p. 031005, 2018.

[17] R. Alomari, M. V. Martin, S. MacDonald, and C. Bellman, "Using EEG to predict and analyze password memorability," in2019 IEEE InternationalConference on Cognitive Computing (ICCC). IEEE, 2019, pp. 42–49.

[18] G.-C. Yang, "Next-generation personal authentication scheme based on EEG signal and deep learning," Journal of Information Processing Systems, vol. 16, no. 5, pp. 1034–1047, 2020.


Related Articles