Systems, Man & Cybernetics - October 2017 - 11

presence of repeatable discriminative characteristics in
the individuals' VEP activity. The authors of [49] investigated the intersession stability of VEPs during two sessions separated by one week and reported an average
EER of 14% with spectral features extracted from six EEG
channels. The study described in [50] provides a highsecurity EEG-based log-in system with a rapid serial visual presentation (RSVP) paradigm. The RSVP paradigm
allows presentation of a large number of stimuli within a
short time, and this study reported a 100% true acceptance
rate using 31 EEG channels among 29 subjects. However,
it is worth noting that the literature does not provide indepth information about the effectiveness and permanence of VEP-based biometric recognition systems except
for the results obtained from relatively small databases
(a maximum of 70 subjects [51]) and across a relatively
smaller time span of six months [31], [32].
Mental Tasks
A variety of brain patterns related to mental activities
have also been explored in biometric recognition [34]-
[37], and the study conducted in [8] reports the usability
of various mental tasks. The five mental activities used
included baseline activity (relaxed and thinking of
nothing in particular), math operation, geometric figure
rotation activity, letter composition, and visual counting. Math operation was reported to be the best task for
distinguishing among subjects. A perfect classification
rate wa s achieved by concatenating the features
extracted from all of the five mental tasks. The studies
in [14] and [34] used motor imagery patterns in a biometric system, and the authors highlighted that lefthand motor imagery was more suitable than that of the
right hand. The authors of [37] conducted a study using
six volunteers to assess the feasibility of using imagined speech for subject identification. Using AR features extracted from 128 channels, subjects were

identifiable at a CRR of 99.76%. The results are promising and highlight the feasibility of imagined speechrelated EEG for biometric recognition. Further investigation
is sparsely reported in the literature, and future research is
thus needed.
Recent developments since 2013 are included in Table 2.
The study in [35] investigates EEG during motor imagery
tasks for person authentication and shows an EER of
0.044% among nine subjects; an increased data set of 109
subjects in [54] offers an HTER of 3% using PSD. Although
Table 2 shows that mental-activity-based EEG patterns are
also capable of providing good biometric performance,
their stability analyses have not yet been reported and
should be exploited using a larger population of subjects
(more than 109).
In addition to the three aforementioned EEG paradigms, a number of independent studies are also reported
in the literature. The studies in [38] and [39] present preliminary investigation results on the feasibility of using
eye blinks captured from frontal EEG electrodes for personal recognition. The work described in [40] and [41] suggests the applicability of using the event-related potential
P300 in biometric recognition systems. P300 is a positive
peak deflection in EEG that occurs at approximately 300 ms
in response to a task-relevant stimulus, and it is mainly
reflected in central and parietal EEG channels. The possibility of using auditory-evoked potential as an EEG biometric feature was also mentioned for the first time in [10].
Future Directions and Challenges
In any experimental protocol adopted for EEGBS, the proper selection of discriminative EEG features is a crucial
determinant in identification or authentication success
rates. It is well known that there is a background noise in
EEG, superimposed to the signals representing the synchronous firing of specific collections of neurons related to
a particular task. Biological artifacts related to eye

Table 2. Recently reported biometric systems using mental tasks.
Reference/Year

Channels/Subjects

Mental Activity

Feature Extraction

Performance

[35], 2013

3/9

Motor imagery of left and
right hands

AR and PSD

EER = 0.044

[52], 2013

3/9

Motor imagery of tongue
and foot

AR and PSD

EER = 0.003

[53], 2015

14/6

Mental writing of the
user-owned password

Multivariate AR model
parameters and
PSD features

HTER = 3-5%

[54], 2016

32/109

Imagination of opening
and closing fists or feet

Binary Flower Pollination
Algorithm

CRR = 87%

[55], 2016

1/5

Motor movement and
imaginary cognitive
process

Wavelet-based features

TAR = 95%
FRR = 4.44%

TAR: true acceptance rate.

O c tob e r 2017

IEEE SYSTEMS, MAN, & CYBERNETICS MAGAZINE

Table of Contents for the Digital Edition of Systems, Man & Cybernetics - October 2017