A Pathway To Implement

Security System Using EEG

Signals of Digit Recognition
Activity.
Rakesh Shamrao Deore1*, M.V.Amrutsagar1, S.C.Mehrotra2
1

Department of Computer Science, SSVPS Science College, Dhule

{rakeshsdeore,}@gmail.com
2
Department of Computer Science and IT,BAMU, Aurangabad
sureshchandmehrotra16.yahoo.com

Abstract. This paper discusses the application of EEG signal in security systems. EEG
signal vary from person to person. It is not possible to steal or replicate the EEG signal.
So the EEG signal has biometric potential. In this study we explored the characteristics
of multichannel Electroencephalogram (EEG), which is recorded from no of subjects
recognizing different numbers displayed on the screen by GUI software designed in
VB6.We processed these EEG signals by using Linear Discriminate Analysis (LDA).
We found that the scaling exponent of each digit is different especially at positions C3
and C4, and at positions O1 and O2. This study shows that it is possible to distinguish
EEG signal of different digits. The results of this paper are used in designing Security
systems based on EEG signal.

Keywords: Brain Computer Interface, EEG, Linear Discriminate Analysis,

Biometric, Security System, Information Security

Introduction

The Electroencephalogram (EEG) is a unique and valuable measure of brains

electrical function. Electrical Signals are produced by brain activity were first
recorded from the cortical surface in animals by Ricard Carton in 1875 (Car-ton 1875)
and from human scalp by Hans Beger in 1929 (Berger,1929)[6]. EEG activities
mainly used for clinical diagnosis and explores brain function. Now a day different
researchers shows that brain activity might serve an entirely different purpose that
they might provide the brain with another means of conveying messages and
commands to external world. EEG is one such technique which measures the electric
fields that are produced by the activity in the brain [4, 10]. EEG signals arise due to
electrical potential produced by the brain. EEG spectrum contain characteristic
waveforms which fall in 4 frequency bands viz alpha (8-13Hz), beta (13-30 Hz), theta
(4-8 Hz), delta (< than 4 Hz). Alpha waves are found in normal awake people, not
engaged in intense mental activity, which disappear when a person is asleep. Beta
1

waves with higher frequency are seen during intense mental activity and stress. Delta
waves occur during deep sleep, during infancy and in serious organic brain diseases.
Theta waves appear during emotional stress in adults in sleep, particularly during
disappointment and frustration .Literature review of similar work which focuses more
on identification from EEG of healthy subjects rather than classification of
pathological cases for diagnosis were made [1,2, 3, 7, 8, 9, 10] EEG is collected at the
millisecond level, in contrast to the longer time intervals required for traditional
measures such as mouse clicks or user responses. This permits effective monitoring of
workload fluctuations in very rapid decision-making processes that are unobservable
using traditional methods [11]. As security issue is always challenging to the real
world applications many biometric approaches, such as fingerprint, iris and retina,
have been proposed to improve recognizing accuracy or practical facility in individual
identification in security. However, there is little research on individual identification
using EEG methodology mainly because of the complexity of EEG signal collection
and analysis in practice [5]. In future we can work to extract individual specific
information from a persons EEG and use this information to develop identification
methods like EEG biometry .The remaining sections are organized as follows. Section
2 provides experimental data acquisition setup used in this research work. Section 3
details the experimental analysis of EEG data, followed by results and conclusion in
section 4.

EXPERIMENTAL DATA ACQUISITION

EEG recordings of 10 male right-handed subjects in the age group of (20-25) were
taken. The subjects were normal without any mental disorder. They did not have any
problem in communicating and had normal vision.

Fig 1. Snapshot of Graphical User Interface Designed in VB6

The subjects had given their written consent for recording EEG signals before
participating. For the EEG data, experiments were conducted in which subjects were
shown visual stimuli consisting of the Random Number Generator. Graphics User
Interface was developed in VB6 .The snapshot of GUI is as shown in fig 1.This GUI
was shown for one minute and EEG signal of corresponding activity was recorded. A
single experimental session typically comprised of 5 trials of complete display of GUI
from 0 to 9. The record of the displayed number was maintained for reference during
analysis and pattern matching. The electrodes are placed using 10-20 system. Here we
study the EEG signals of six Electrodes [C3, C4, P3, P4, O1, O2] as shown in gure
2.

Fig 2. 10-20 system for Electrode Placement.

As the architecture of the brain varies with different locations, EEGs can vary
depending on the location of recording electrodes. Electrodes consist of at discs
connected to an isolated wire. They have identifying names; those on the left side
have odd numbers, while those on the right have even numbers. Those near the
midline have smaller number and the more lateral ones have larger numbers. The
names include the rst letter of the place where the electrode is placed. A body earth
and reference electrode was placed on forehead and behind the two ear lobes.

EXPERIMENTAL ANALYSES

The EEG recordings were captured according to the time of numbers generated on
the GUI. The readings were statistically analyzed. There are four frequency bands
associated with the EEG signals. All fours bands with their functionalities are listed in
table 1[7].Type Frequency (Hz) Normally Delta up to 4 Has been found dominant
during some continuous attention tasks.

Table1:- Frequency Banda and Their Functionality.

Name of Frequency
Range
Activity
Band
Delta

Upto 4

Theta

4-<8

Alpha

8-13

Beta

13-30

Has been found dominant during some

continuous attention task
Appears in drowsiness or arousal
condition in children and adults
Found to be prominent in relaxed
reflecting condition
Has been found in alert working active,
busy or anxious thinking, or active
concentration.

By observing the above table, we decided to concentrate on the beta signals of all 6
electrodes for recognizing the number. The Linear Discriminate analysis (LDA) is
used for data classification and dimensionality reduction. This method easily handles
the case where the Within-class frequencies are unequal and their performance has
been examined on randomly generated test data. This method maximizes the ratio of
between-class variance to the within-class variance in any particular data set thereby
guaranteeing maximal severability.
3. 1 Linear Discriminant Analysis
Linear Discriminate Analysis is a well-known scheme for feature extraction and
dimension reduction. It has been widely used in many application such as face
recognition, image retrieval, microarray classication. LDA projects the data onto a
lower dimensional vector space such that the ratio of the between class distance to the
within class distance is maximized, thus achieves maximum discrimination
In this section we give a brief overview of LDA. Given a matrix A RNxn,LDA aims
to nd a transformation G RNxl that maps each column a i of A, for 1 i n,in the
N-dimensional space to a vector bi in the l-dimensional space. That is G : a i RN >
bi = GT ai Rl (l < N). LDA aims to nd a vector space G spanned by g i where G =
[g1, g2, ..., gl] such that each ai is projected onto G by (g1T .ai, ..., glT .ai)T Rl.
Assume that the original data in A partitioned into k classes as A={1, 2, ..., k}
k

where i contains ni points from the ith class, and

=n

.LDA aims to nd the

i=1

optimal transformation G such that the class structure of the original high dimensional
space is preserved in the low dimensional space. In general if each class is tightly
grouped, but well separated from the other classes , the quality of considered to be
high. In discriminate analysis ,two scatter matrices, called as within class (Sw) and
between-class (Sb) matrices, are dened to quantify the quality of cluster as follows:
Sw = i = 1k x i (x - mi) (x - mi)T
Sb = i = 1k ni ( mi - m) ( mi - m)T

Where mi = 1/ni x i, x is the mean of the ith class and m=1/n i = 1 k x

i, x is the global mean. It is easy to verify that trace (Sw) measures the closeness of
the vectors within the classes, while trace (Sb) measures the separation between
classes. In the low dimensional space resulting from the linear transformation G.
Following fig 3. shows the classification between all ten numbers using LDA.

of

Fig3.
to 9

Classification of EEG data

digits using LDA

Table
2
information
X
0.009345304
0.004600992
0.002613943
-0.015974174
0.002046017
-0.012309459
-0.001831391
-0.00399276
0.008251523
-0.008728191

Cluster
for digit-0
Y
0.146554217
0.190279302
0.119080978
0.126782597
0.143718024
0.118067418
0.16674048
0.175422066
0.131259941
0.140959334

Distance
0.00173175
0.044421612
0.026853037
0.019170542
0.002693518
0.027864866
0.020915166
0.029578819
0.01471351
0.005179706

Table 3 Cluster information for digit-1

X
Y
Distance
0.10556187
0.092780159
0.892034621
0.108956621
0.098769387
0.108096731
0.102586447
0.102800257
0.108370003
0.109214756
0.094587284
0.107971648
0.110702266
0.080517333
0.108737101
0.110771811
0.091228471
0.107988423
Table 2 and Table 3 shows the cluster information of EEG signal related to digit 0 and
1.Here we only list the cluster information for digits 0 and 1 as sample information.

We process each cluster as follows to calculate center and radius of each cluster. The
resultant information is given in table 4.
All data points in a single cluster are called as cluster feature (CF).Let CF is a data
structure summarizing information about all points in the dataset,
CF = (n,LS)
where LS is linear sum of n data points.
ie
n

xi
i=1

yi

i=1

n is the number of data points in data set.

Mean of data set provides the center of (x 0,y0) of the distribution. Then the distance
of each point in a cluster from the center of cluster is given by

( xixo ) 2+ ( yi y 0 ) 2
The radius of cluster is calculated as
n

( xixo ) 2+( yi y 0 ) 2/

k=0

Table 4. Center and Radius of clusters of different Digits

Digit
0
1
2
3
4
5
6
7
8
9

X
-0.0016
0.107965628
-0.064113698
0.032867432
0.106169638
0.10241185
-0.278322
-0.006807753
0.045618502
-0.035999385

Y
0.145886
0.093447148
0.096575553
-0.22817591
0.034016434
-0.375910337
-0.140508139
0.287239623
0.013377525
0.122955778

Fig 4 Snapshot of Clusters for different digits.

Distance
0.019312253
0.238866421
0.065449459
0.288061438
0.109176504
0.103655998
0.279101655
0.157660533
0.048312754
0.042695393

Result and Conclusion

Linear Discriminate Analysis is a well-known scheme for feature extraction and

dimension reduction. Recording are sampled at 250 Hz for 90 sec so length of data is
very large. But it is transformed using fourier transformation into frequency data. For
each digit we get 6 scaling exponent responding each channel. The data is ltered
by band pass lter. The effective component of EEG is usually supposed to be
concentrated at 1-30 HZ and frequency of baseline excursion is comparatively lower (
1 HZ) and the frequency of power interruption and EMG interruption is
higher(>=50HZ).so we set the parameters as Low lter 1 Hz, High lter 70 Hz
,Number of channels:06,Sweep speed 30 mm/s and Montage: BP PARA
(R).Considering the calculation results for the same digit, the results of one subject
during each trail are comparatively consistent. Table 2 to 3 displays the data points of
the clusters for digits 0 to 1.Table 4 displays the center and radius for digits 0 to 9.The
frequency is different for different digits. There is well separation between the
clusters of digits. In this study ,we proposed an effective method for extracting
properties of EEG data. This study is used to effectively extract EEG features mainly
we concentrated on beta frequency bands. In earlier studies we were concentrated on
19 electrodes. Here now we actually concentrate on 6 electrodes and we get excellent
clusters for the digit recognition. Thus our study is foundation for using Brain
Computer Interface in security systems

References
1.

A.Kostov. Parallel man-machine training in development of eeg-based cursor

control. Proc IEEE,8(6):203205, 2000.
2. A.Kostov. Parallel man-machine training in development of eeg-based cursor
control. Proc IEEE,8(9):203205, 2000.
3. F.Babilioni et.al. Linear classication of low-resolution EEG patterns
produced by imagined hand movement.Proc IEEE, 8(8):186188, 2000.
4.
Muhammad Kamil Abdullah et.al. Analysis of eeg signal for practical
biometric system. World academy of science, Engineering and
Technology,68(2):11231127, 2010.
5. Qinglin Zhao et.al. Improving individual identication in security check with
an eeg. School of information science and engineering Lanzhou
universityLanzhou,(11), 2010.
6. H.Berger. Under das electrenkephalogramm desmenchen. Arch Psychiat
Nervenkr, 87(1):527570,1929.
7 J.J.Vidal. Real time detetction of brain events in eeg.Proc IEEE, 65(4):633664,
1977.
8. J.R.Wolpaw. Brain-computer interface research at the wanderworth center. Proc
7

IEEE, 8(7):222226, 2000.

9. P.R.Kennedy. Direct control of computer from the human central nervous sy
tem. Proc IEEE,8(5):198202, 2000.
10. Shashibala Rao PriyankaAbhang S.C.Mehrotra. Region wise emotion recogni
tion using eeg brain signals. Proc. International conference on Machine Intelli
gence Application to Power, Signal Processing,Communication and Control,
65(3):11231127, 2010.
11. R. et.al. Stevens. Allocation of time eeg-engagement and eeg-workload re
sources as scientic problem solving skills are acquired in the classroom. Aug
mented cognition: Past, present, and future. Strategic Analysis, (10), 2002.