EEG-based Biometric Identification Research Articles

Investigation of EEG-Based Biometric Identification Using State-of-the-Art Neural Architectures on a Real-Time Raspberry Pi-Based System

Despite the growing interest in the use of electroencephalogram (EEG) signals as a potential biometric for subject identification and the recent advances in the use of deep learning (DL) models to study neurological signals, such as electrocardiogram (ECG), electroencephalogram (EEG), electroretinogram (ERG), and electromyogram (EMG), there has been a lack of exploration in the use of state-of-the-art DL models for EEG-based subject identification tasks owing to the high variability in EEG features across sessions for an individual subject. In this paper, we explore the use of state-of-the-art DL models such as ResNet, Inception, and EEGNet to realize EEG-based biometrics on the BED dataset, which contains EEG recordings from 21 individuals. We obtain promising results with an accuracy of 63.21%, 70.18%, and 86.74% for Resnet, Inception, and EEGNet, respectively, while the previous best effort reported accuracy of 83.51%. We also demonstrate the capabilities of these models to perform EEG biometric tasks in real-time by developing a portable, low-cost, real-time Raspberry Pi-based system that integrates all the necessary steps of subject identification from the acquisition of the EEG signals to the prediction of identity while other existing systems incorporate only parts of the whole system.

Person-identification using familiar-name auditory evoked potentials from frontal EEG electrodes

Electroencephalograph (EEG) based biometric identification has recently gained increased attention of researchers. However, state-of-the-art EEG-based biometric identification techniques use large number of EEG electrodes, which poses user inconvenience and consumes longer preparation time for practical applications. This work proposes a novel EEG-based biometric identification technique using auditory evoked potentials (AEPs) acquired from two EEG electrodes. The proposed method employs single-trial familiar-name AEPs extracted from the frontal electrodes Fp1 and F7, which facilitates faster and user-convenient data acquisition. The EEG signals recorded from twenty healthy individuals during four experiment trials are used in this study. Different combinations of well-known neural network architectures are used for feature extraction and classification. The cascaded combinations of 1D-convolutional neural networks (1D-CNN) with long short-term memory (LSTM) and with gated recurrent unit (GRU) networks gave the person identification accuracies above 99 %. 1D-convolutional, LSTM network achieves the highest person identification accuracy of 99.53 % and a half total error rate (HTER) of 0.24 % using AEP signals from the two frontal electrodes. With the AEP signals from the single electrode Fp1, the same network achieves a person identification accuracy of 96.93 %. The use of familiar-name AEPs from frontal EEG electrodes that facilitates user convenient data acquisition with shorter preparation time is the novelty of this work.

EEG‐based biometric identification using frequency‐weighted power feature

The brain activity of a person sensed using electroencephalograph (EEG) has a unique neuronal signature consisting of relevant subject-specific information, which can be used for biometric identification. The main challenge in brainwave-based biometric identification is the extraction of robust features from non-stationary EEG signal that provide sufficient discriminability to differentiate between individuals. In this study, the authors propose an EEG-based biometric identification technique using a novel feature called frequency-weighted power (FWP), which offers higher discrimination in the person identification compared to the state of the art EEG features. FWP is an equivalent representation of the power of a specific frequency band, obtained by multiplying the specific frequency with its corresponding power density value and summing up it over the specific band. The efficacy of the proposed method is validated using resting-state EEG from online PhysioNet database as well as using resting-state EEG data acquired from 16 subjects in the laboratory during the experiment. Using a correlation-based classifier, the proposed method achieves an equal error rate (EER) of 0.0039 from eyes-closed resting-state EEG signals using 20 electrodes, which is nearly one-fifth of the EER obtained by the best method reported in the literature for the comparable number of electrodes.

BrainPrint: EEG biometric identification based on analyzing brain connectivity graphs

Research on brain biometrics using electroencephalographic (EEG) signals has received increasing attentions in recent years. In particular, it has been recognized that the brain functional connectivity reflects individual variability. However, many questions need to be answered before we can properly use distinctive characteristics of brain connectivity for biometric applications. This paper proposes a graph-based method for EEG biometric identification. It consists of a network estimation module to generate brain connectivity networks and a graph analysis module to generate topological features based on brain networks. Specifically, we investigate seven different connectivity metrics for the network estimation module, each of which is characterized by a certain signal interaction mechanism, defining a peculiar subjective brain network. A new connectivity metric is proposed based on the algorithmic complexity of EEG signals from a information-theoretic perspective. Meanwhile, six nodal features and six global features are proposed and studied for the graph analysis module. A comprehensive evaluation is carried out to assess the impact of different connectivity metrics, graph features, and EEG frequency bands on biometric identification performance. The results demonstrate that the graph-based method proposed in this study is effective in improving the recognition rate and inter-state stability of EEG-based biometric identification systems. Our findings about the network patterns and graph features bring a further understanding of distinctiveness of humans’ EEG functional connectivity and provide useful guidance for the design of graph-based EEG biometric systems.

EEG Based Biometric Identification Using Correlation and MLPNN Models

<p class="0abstract">This study investigates the capability of electroencephalogram (EEG) signals to be used for biometric identification. In the context of biometric, recently, researchers have been focusing more on biomedical signals to substitute the biometric modalities that are being used nowadays as the signals obtained from our bodies is considered more secure and privacy-compliant. The EEG signals of 6 subjects were collected where the subjects were required to undergo two baseline experiments which are, eyes open (EO) and eyes closed (EC). The signals were processed using a 2nd order Butterworth filter to eliminate the unwanted noise in the signals. Then, Daubechies (db8) wavelet was applied to the signals in the feature extraction stage and from there, Power Spectral Density (PSD) of alpha and beta waves was computed. Finally, the correlation model and Multilayer Perceptron Neural Network (MLPNN) was applied to classify the EEG signals of each subject. Correlation model has yielded great significant difference of coefficient between autocorrelation and cross-correlation where it gives the coefficient value of 1 for autocorrelation and the coefficient value of less than 0.35 for cross-correlation. On the other hand, the MLPNN model gives an accuracy of 75.8% and 71.5% for classification during EO and EC baseline condition respectively. Therefore, these results support the usability of EEG signals in biometric recognition.</p>

Non Linear Features Analysis between Imaginary and Non-imaginary Tasks for Human EEG-based Biometric Identification

Electroencephalogram (EEG) is a signal contains information of brain activities. Nowadays, many types of research regarding EEG have been done such as neuromarketing. The human brain is very complicated but it contains a lot of information. EEG signal is a non-stationary signal, it changes over time and it also depends on human’s emotion, thinking and activities. Due to the uniqueness of the EEG signal, the EEG signal has the potential to be used in the human authentication system. In this paper, an imaginary task and a non-imaginary task were studied to find out which type of task is possible to be used in authentication system. In preliminary study, five subjects were volunteered and performed the motor imagery and motor execution tasks. EEGOTM sports device (ANT Neuro, Enschede, Netherlands) with 32 channels was used to record the EEG signal and the sampling frequency is set to 512 Hz. The EEG signal was analysed by using EEG signal processing namely pre-processing, feature extraction and classification. Power line interference was removed by using a notch filter. Daubechies 8 wavelet family with 5th level decomposition has been applied to remove baseline wander noise. The performance of non-linear features such as Empirical Mode Decomposition (EMD), Hurst Exponent and Entropy were examined. Random forest gives good classification accuracy for imaginary task and non-imaginary task which are 83.53% and 87.06% respectively, thus, it shows non-linear features is possible to be employed in biometric identification.

EEG-based biometric identification with convolutional neural network

Although more interest arising in biometric identification with electroencephalogram (EEG) signals, there is still a lack of simple and robust models that can be applied in real applications. This work proposes a new convolutional neural network with global spatial and local temporal filter called (GSLT-CNN), which works directly with raw EEG data, not requiring the need for engineering features. We investigate the performance of the GSLT-CNN model on datasets of 157 subjects collected from 4 different experiments that measure endogenous brain states (driving fatigue and emotion) as well as time-locked artificially induced brain responses such as rapid serial visual response (RSVP). We evaluate the GSLT-CNN model against the comparable SVM, Bagging Tree and LDA models with effective feature selection method. The results show the GSLT-CNN model is highly efficient and robust in training more than 279 K epochs within less than 0.5 h and achieves 96% accuracy in identifying 157 subjects, which is 3% better than the best accuracy of SVM on selected PSD feature, 10% better than that of SVM on selected AR feature and 23% better than that of normal CV-CNN model on raw EEG feature. It demonstrates the potential of deep learning solutions for real-life EEG-based biometric identification. We also show that the cross-session identification accuracy from time-locked RSVP data (99%) is slightly higher than that from single-session non-time-locked driving fatigue data (97%) and much higher than that from epochs measuring random brain states (90%), which implies RSVP could be a more beneficial design to achieve high identification accuracy with EEG and our GSLT-CNN model is robust for cross-session identification in RSVP experiment.

EEG Fingerprints: Phase Synchronization of EEG Signals as Biomarker for Subject Identification

The goal of biometrics is to recognize humans based on their physical and behavioral characteristics. Preliminary studies have demonstrated that the electroencephalogram(EEG) is potentially more secure and private than traditional biometric identifiers. At present, the EEG identification method targets specific tasks and cannot be generalized. In this study, a novel EEG-based biometric identification method that extracts the phase synchronization (PS) features for subject identification is proposed under a variety of tasks. We quantified the PS features by the phase locking value (PLV) in different frequency bands. Subsequently, we employed the principal component analysis (PCA) to reduce the dimension. Then, we used the linear discriminant analysis (LDA) to construct a projection space and projected the features onto the projection space. Finally, a feature vector was assigned to the class label. The experimental results of the proposed method used on 3 datasets with different cognitive tasks showed high classification accuracies and relatively good stabilities. From the results, we found that particularly in the beta and gamma bands, the average accuracies are more than 97% with the standard deviation equal to or less than the magnitude 10e-2 for both Dataset 1 and Dataset 2. For Dataset 3, the PS feature vectors in all off the bands have high classification accuracies, which are more than 97% with the standard deviation of the same magnitude. Our work demonstrated that the phase synchronization of EEG signals has task-free biometric properties, which can be used for subject identification.

MindID

Person identification technology recognizes individuals by exploiting their unique, measurable physiological and behavioral characteristics. However, the state-of-the-art person identification systems have been shown to be vulnerable, e.g., anti-surveillance prosthetic masks can thwart face recognition, contact lenses can trick iris recognition, vocoder can compromise voice identification and fingerprint films can deceive fingerprint sensors. EEG (Electroencephalography)-based identification, which utilizes the user's brainwave signals for identification and offers a more resilient solution, has recently drawn a lot of attention. However, the state-of-the-art systems cannot achieve similar accuracy as the aforementioned methods. We propose MindID, an EEG-based biometric identification approach, with the aim of achieving high accuracy and robust performance. At first, the EEG data patterns are analyzed and the results show that the Delta pattern contains the most distinctive information for user identification. Next, the decomposed Delta signals are fed into an attention-based Encoder-Decoder RNNs (Recurrent Neural Networks) structure which assigns varying attention weights to different EEG channels based on their importance. The discriminative representations learned from the attention-based RNN are used to identify the user through a boosting classifier. The proposed approach is evaluated over 3 datasets (two local and one public). One local dataset (EID-M) is used for performance assessment and the results illustrate that our model achieves an accuracy of 0.982 and significantly outperforms the state-of-the-art and relevant baselines. The second local dataset (EID-S) and a public dataset (EEG-S) are utilized to demonstrate the robustness and adaptability, respectively. The results indicate that the proposed approach has the potential to be widely deployed in practical settings.

On the invariance of EEG-based signatures of individuality with application in biometric identification.

One of the main challenges in EEG-based biometric systems is to extract reliable signatures of individuality from recorded EEG data that are also invariant against time. In this paper, we investigate the invariability of features that are extracted based on the spatial distribution of the spectral power of EEG data corresponding to 2-second eyes-closed resting-state (ECRS) recording, in different scenarios. Eyes-closed resting-state EEG signals in 4 healthy adults are recorded in two different sessions with an interval of at least one week between sessions. The performance in terms of correct recognition rate (CRR) is examined when the training and testing datasets are chosen from the same recording session, and when the training and testing datasets are chosen from different sessions. It is shown that an CRR of 92% can be achieved based on the proposed features when the training and testing datasets are taken from different sessions. To reduce the number of recording channels, principal component analysis (PCA) is also employed to identify channels that carry the most discriminatory information across individuals. High CRR is obtained based on the data from channels mostly covering the occipital region. The results suggest that features based on the spatial distribution of the spectral power of the short-time (e.g. 2 seconds) ECRS recordings can have great potentials in EEG-based biometric identification systems.