Steady-State Visual Evoked Potentials for EEG-Based Biometric Identification

On the Influence of Affect in EEG-Based Subject Identification

“…It should be noted that we do not attempt to build an EEG-based biometric system, which has been the subject of many other previous works in the literature [43], [44], [45]. Rather, we aim to determine whether the user's affective state may alter the user identification performance of an EEG-based biometric system.…”

“…The 20 000 test sub-sets previously created for the SEED dataset were used in order to repeat the SAME and DIFF experiments as previously described, with the additional constraint that the training sets for the classification models could only contain samples referring to a different session than the ones contained in the test set. Furthermore, apart from the PSD features, we also examined the performance of the Mel Frequency Cepstral Coefficients (MFCC) features and the Auto Regression Reflection Coefficients (ARRC) features, as proposed by Piciucco et al [45] for EEG-based subject identification. As also proposed in [45], the samples were divided into frames with a length of 5 sec and 75% overlapping, leading to a total of 46 new samples generated from each of the original recordings in the dataset.…”

“…Furthermore, apart from the PSD features, we also examined the performance of the Mel Frequency Cepstral Coefficients (MFCC) features and the Auto Regression Reflection Coefficients (ARRC) features, as proposed by Piciucco et al [45] for EEG-based subject identification. As also proposed in [45], the samples were divided into frames with a length of 5 sec and 75% overlapping, leading to a total of 46 new samples generated from each of the original recordings in the dataset. Then, the 20 000 sub-sets were computed again from the newly created samples of the SEED dataset and the SAME and DIFF experiments were conducted again for the MFCC and the ARRC features.…”

Image-Evoked Affect and its Impact on Eeg-Based Biometrics

“…MFCCs are a parametric representation of the Fourier Spectrum and have been widely used in speech recognition [22,23] and more recently applied to EEG-based subject identification [24,25], achieving relatively high accuracies. In this work we use the well-known MFCC computation using HTK-like filterbanks and Discrete Cosine Transform.…”

“…In this work we use the well-known MFCC computation using HTK-like filterbanks and Discrete Cosine Transform. Following [24], MFCC features have been computed using 18 filterbanks, producing a total 12 cepstral coefficients per channel. The final feature vector is created by concatenating the cepstral coefficients of each of the channels, leading to a total of 168 features (12 coefficients × 14 channels).…”