EEG-based biometric recognition using EigenBrains

Abstract

An increased level of attention has recently raised on biometric recognition by means of electroencephalography (EEG). This modality in fact possesses several properties which may be appealing for automatic people recognition, such as the intrinsic liveness detection and the robustness against potential attacks. Moreover, it could be easily exploited in applications based on brain-computer interfaces (BCI). In this paper we exhaustively analyze the discriminative capability of a compact representation of EEG signals acquired in resting conditions. Specifically, the exploited templates are obtained as projections into a subspace defined through EigenBrains (EBs), a basis for EEG data relying on principal component analysis (PCA). An extensive set of experimental tests, conducted on a database comprising 60 users, is performed to evaluate the recognition capabilities of the proposed representation under different system configurations

A Survey on Brain Biometrics

“…Falzon et al [36] used 15 principle components to retain 90% of the total variance. PCA and Multilinear PCA (MPCA) were used in brain biometric recognition [95,96], as well as for PSD feature projection (EB and ETB) [96].…”

“…If the similarity metric was not explicitly specified in the paper, it was marked as "kNN." Commonly used similarity measurements, like Cross-Correlation (CC) [6, 142-146, 167, 168, 172, 173], Cosine Similarity [28,32,62,[95][96][97]106], Euclidean Distance [9, 10, 16, 39, 42, 44, 64, 95-97, 109, 121, 124, 174, 181], Mahalanobis Distance [81,95,96,166,167], Manhattan Distance [54, 95-97, 117, 119, 121, 130, 136], and Bhattacharyya Distance [83] have been investigated in the literature.…”

“…, 41 Hz[95,96], and 50 Hz[81], or bandpass filters of 0.01-50 Hz[27], 0.01-100 Hz [31], 0.1-42 Hz [172], 0.1-55 Hz [98], 0.1-100 Hz [36], 0.5-8 Hz [33], 0.5-30 Hz [78, 79], 0.5-40 Hz [114, 130], 0.5-45 Hz [168], 0.5-50 Hz [9], 0.5-70 Hz [138, 147], 1-30 Hz [133], 1-50 Hz [143], 2-47 Hz [74], 2-50 Hz [65, 83], 2-55 Hz [64], 4-52 Hz [4], 5-100 Hz [50, 51], 8-30 Hz [52, 54, 92, 162], 14-50 Hz [53], 20-50 Hz…”

Private authentication keys based on wearable device EEG recordings

“…When using 2 to 4 channels on the eyes open/eyes closed modality, the highest accuracy achieved was 81%. Recently, Campisi et al [6] proposed a new method, called Eigenbrain, that used VEP to realize an EEG-based authentication system. There the EEG signals were modeled using principal component analysis (PCA) applied to the EEG power spectrum density (PSD) over 19 electrodes, and classification was done using linear discriminant analysis (LDA).…”