J. Biomedical Science and Engineering, 2010, 3, 390-396 JBiSE
doi:10.4236/jbise.2010.34054 Published Online April 2010 (http://www.SciRP.org/journal/jbise/).
Published Online April 2010 in SciRes. http://www.scirp.org/journal/jbise
Classification of human emotion from EEG using discrete
wavelet transform
Murugappan Murugappan, Nagarajan Ramachandran, Yaacob Sazali
School of Mechatronic Engineering, Universiti Malaysia Perlis (UniMAP), Perlis, Malaysia.
Email: murugappan@unimap.edu.my
Received 21 November 2009; revised 28 December 2009; accepted 4 February 2010.
In this paper, we summarize the human emotion rec-
ognition using different set of electroencephalogram
(EEG) channels using discrete wavelet transform. An
audio-visual induction based protocol has been de-
signed with more dynamic emotional content for in-
ducing discrete emotions (disgust, happy, surprise,
fear and neutral). EEG signals are collected using 64
electrodes from 20 subjects and are placed over the
entire scalp using International 10-10 system. The
raw EEG signals are preprocessed using Surface
Laplacian (SL) filtering method and decomposed into
three different frequency bands (alpha, beta and
gamma) using Discrete Wavelet Transform (DWT).
We have used “db4” wavelet function for deriving a
set of conventional and modified energy based fea-
tures from the EEG signals for classifying emotions.
Two simple pattern classification methods, K Nearest
Neighbor (KNN) and Linear Discriminant Analysis
(LDA) methods are used and their performances are
compared for emotional states classification. The ex-
perimental results indicate that, one of the proposed
features (ALREE) gives the maximum average classi-
fication rate of 83.26% using KNN and 75.21% using
LDA compared to those of conventional features. Fi-
nally, we present the average classification rate and
subsets of emotions classification rate of these two
different classifiers for justifying the performance of
our emotion recognition system.
Keywords: EEG; Human Emotions; Discrete Wavelet
Transform; KNN; LDA
Emotion is one of the most important features of humans.
Without the ability of emotions processing, computers
and robots cannot communicate with human in natural
way. It is therefore expected that computers and robots
should process emotion and interact with human users in
a natural way. In recent years, research efforts in Human
Computer Interaction (HCI) are focused on the means to
empower computers to understand human emotions. Al-
though limited in number compared with the efforts be-
ing made towards intention-translation means, some re-
searchers are trying to realize man-machine interfaces
with an emotion understanding capability. Most of them
are focused on facial expression recognition and speech
signal analysis [1,2]. In these works, the emotions of the
subjects are purposefully expressed and it can be more
easily concealed by those of other subjects [3]. Another
possible approach for emotion recognition is physio-
logical signal analysis. We believe that this is a more
natural means of emotions recognition, in that the influ-
ence of emotion on facial expression or speech can be
suppressed relatively easily, and emotional status is in-
herently reflected in the activity of nervous system. The
traditional tools for the investigation of human emo-
tional status are based on the recording and statistical
analysis of physiological signals from the both central
and autonomic nervous systems. Some of the main
physiological signals highly adopted for human emotion
assessment are: Electromyogram (EMG), Electrocardio-
gram (ECG), Skin Conductive Resistance (SCR), and
Blood Volume Pressure (BVP) [4,5]. Several approaches
have been reported by different researchers on finding
the correlation between the emotional changes and EEG
signals [6-8]. The past works on emotion recognition
using EEG signals are reported in [9].
One of the difficulties of categorizing emotion names
is that the distinction between the emotion categories is
not clear. All people express their emotions differently: it
is not an easy task to judge or to model emotions. Re-
searchers often use two different methods to model emo-
tions [10,11]. One approach is word labels, for example,
happy, sad, surprise, anger, fear, disgust, etc. Another
ways is to have multiple dimensions or scales to catego-
rize emotions. Instead of choosing discrete labels, ob-
servers can indicate their impressions of each stimulus on
several continuous scales: for example, pleasant-unpleasant,
attention-rejection, simple-complicated, etc. Many re-
searchers have attempted to utilize the two dimensional
M. Murugappan et al. / J. Biomedical Science and Engineering 3 (2010) 390-396
Copyright © 2010 SciRes. JBiSE
modeling of emotions due to simplicity of protocol de-
sign and signal processing techniques [12,13]. However,
our work is to realize the basic emotions that are in dis-
crete mode for developing intelligent emotion recogni-
tion system. Furthermore, the assessment of discrete mode
emotions is useful to create suitable reactions in man-
machine systems such as robots for disabled. Indeed, the
development of human emotion recognition system is
challenging in terms of developing protocol design and
in terms of developing efficient machine learning algo-
rithms for deriving the discrete emotion from the multi-
ple complex emotions encountered by human.
In order to provide a simplified emotion recognition
system, in our earlier work, we proposed asymmetric
ratios based channel selection for reducing the number
of channels from 62 to 8 and to 4 [14]. Since, the reduc-
tion of channels does minimize the physical burden,
mental fatigue during electrode placement, computa-
tional time and complexity. The basic EEG based emo-
tion recognition system is shown in Figure 1.
The rest of this paper is organized as follows. In Sec-
tion 2, we summarize the research methodology by elu-
cidating the data acquisition process, preprocessing, and
feature extraction using wavelet transform, and classifi-
cation of emotions by linear classifiers. Section 3 illus-
trates the overview of the results and discussion of this
present work, and conclusions are given in Section 4.
2.1. EEG Data Acquisitions
This section describes the acquisition of EEG signals for
emotion assessment experiment. A pilot panel study is
conducted on 25 university students to select any 5 video
clips (trials) for each emotion from 115 emotional video
clips including from the international standard emotional
clips1. All the video clips are short in time duration and with
Figure 1. Basic emotion recognition system using EEG.
more dynamic emotional content. The selection of video
clips is based on self assessment questionnaires men-
tioned in [15]. The subjects who have undergone this
panel study do not take part in the data collection ex-
periment. Three females and seventeen males in the age
group of 21-39 years were employed as subjects in our
experiment. All these subjects are university students
and do not have psychiatric or neurological disorders.
All the subjects have given written consent prior to the
recording. Every participant was given information
about the design and purpose about the experiment. The
subjects were paid and allowed to perform the experi-
ments during their free times during working days
(Monday ~ Friday).
The EEG signals were acquired using the Nevus EEG,
Iceland with 64 electrodes. There are 62 active elec-
trodes plus one electrode for reference and one electrode
for ground (Figure 2(a)). All these electrodes are placed
on the scalp according to International 10-10 system and
made up of Ag/Ag-Cl. The reference electrode AFz is
placed in between AF1 and AF2 electrodes and the ground
electrode Oz is placed in between O1 and O2 electrodes.
The impedances of the electrodes were kept below 5K.
The EEG signals collected from the scalp electrodes are
filtered on-line (band-pass) between 0.5 Hz and 70 Hz.
The sampling rate was fixed at 256 samples per second
for all the channels and digitized at 16 bit. The recording
of EEG signals for various emotional stimuli is carried
out in an electrical and sound attenuated room while the
participant was seated comfortably on the armchair.
2.2. Emotion Elicitation Method
Emotions can be induced by one of the following ways:
1) visual (images/pictures) [12]; 2) audio-visual (film
clips/video clips) [13,16]; 3) recalling of past emotional
events [8]; 4) audio (songs/sounds) [17]. Most of re-
searchers are using visual stimuli for evoking emotions.
In our previous work, we have used both visual and au-
dio-visual stimuli for evoking discrete emotions. The
result of this study confirms that, audio-visual stimulus
performs superior in evoking emotions than visual
stimulus [18].
The main advantage of this method resides in the strong
correlation between induced emotional states and the
physiological responses. Hence, we have designed an
audio-visual induction based protocol for eliciting the
discrete emotions in this present work. The complete pro-
tocol design is computerized to follow in prescribed time
period. Thereby no human intervention is required to dis-
turb the experimental session. In order to get quality data,
the subjects are asked to follow a set of instruction prior to
the experimental session: 1) sit back comfortably; 2) do
not move until the end of the experiment session; 3) let
yourself freely experience whatever emotions as much as
you can; 4) do not try to hold back your feelings.
Noise free
(Surface Laplacian
Feature E xtraction
(Wave let Trans-
Emotion Classifi-
(Linear Classifiers)
M. Murugappan et al. / J. Biomedical Science and Engineering 3 (2010) 390-396
Copyright © 2010 SciRes. JBiSE
Figure 2. Electrodes diagram (a) 62 channel; (b) 24
channel; (c) 8 channel.
The audio-visual stimulus protocol for one of the trials
of our experiment is shown in Figure 3. The orders of
the emotional video clips are changed in a random man-
ner for other trials. X1 to X5 denote time periods of se-
lected video clips. The time duration of video clips vary
from one another. In this protocol design, the current
emotional state of the subjects before the experiment
session is neutralized by showing a set of “natural
scenes” and also playing a “soothing music” will make
the subjects to feel calm with relaxed mind. Since, in
most of the experimental sessions, previous researchers
have tolerated the effect of current emotional state of the
subjects before undergoing the emotions assessment
experiment. Between each emotional stimulus (video
clips), the “blank (dark) screen” is displayed for 10 sec-
ond to “rest and prepare” the subject to next stimulus.
Between each emotional video clips, under self assess-
ment section, the subjects were informed to answer the
emotions they have experienced [18]. Finally, 5 trials for
disgust, happy and surprise emotions and 4 trials for fear
and neutral emotions are considered for further analysis.
All the signals are collected without imparting discom-
fort to the subjects.
2.3. Preprocessing
EEG signals recorded over various positions on the scalp
are usually contaminated with noises and artifacts (Ocu-
lar (EoG), Muscular (EMG), Vascular (ECG) and Gloss
kinetic artifacts). The complete removal of artifacts will
also remove some of the useful information of EEG sig-
nals. This is one of the reasons why considerable ex-
perience is required to interpret EEGs clinically [19,20].
A couple of methods are available in the literature to
avoid artifacts in EEG recordings. However, removing
artifacts entirely is impossible in the existing data acqui-
sition process.
In this work, we use of Surface Laplacian (SL) fil-
ter for removing the noises and artifacts. The SL filter is
used to emphasize the electric activities that are spatially
close to a recording electrode (“i”), filtering out those
that might have an origin outside the skull. In addition,
this filter also attenuates the EEG activity which is
Emotion ind ucing video clips
Time in Sec
Blank scree n
Dis Ha
15 10 20 10 X1
2 X3 X4X5
Dis: Disgust Hap: Happy Sur: Surprise Fea: Fear Neu: Neutral
Natural Scene
Figure 3. EEG data acquisition protocol using audio-visual
M. Murugappan et al. / J. Biomedical Science and Engineering 3 (2010) 390-396
Copyright © 2010 SciRes. JBiSE
common to all the involved channels in order to improve
the spatial resolution of the recorded signal. The neural
activities generated by the brain, however, contain vari-
ous spatial frequencies. Potentially useful information
from the middle frequencies may be filtered out by the
analytical Laplacian filters. Hence, the signal “pattern”
derived from SL filters is similar to “spatial distribution
of source in the head”.
The mathematical modeling of Surface Laplacian fil-
ter is given as
() ()
new i
XtX t
where Xnew: filtered signal; X(t): raw signal; NE: number
of neighbor electrodes.
2.4. Feature Extraction
In the emotion recognition research using EEG signals,
the non-parametric method of feature extraction based
on multi-resolution analysis of Wavelet Transform (WT)
is quite new. The joint time-frequency resolution ob-
tained by WT makes it as a good candidate for the ex-
traction of details as well as approximations of the signal
which cannot be obtained either by Fast Fourier Trans-
form (FFT) or by Short Time Fourier Transform (STFT)
[21,22]. The non-stationary nature of EEG signals is to
expand them onto basis functions created by expanding,
contracting and shifting a single prototype function (Ψa,b,
the mother wavelet), specifically selected for the signal
under consideration.
The mother wavelet function Ψa, b (t) is given as
()( )
where a, b R, a > 0, and R is the wavelet space.
Parameters ‘a’ and ‘b’ are the scaling factor and shift-
ing factor respectively. The only limitation for choosing
a prototype function as mother wavelet is to satisfy the
admissibility condition (Eq.3),
where Ψ(ω) is the Fourier transform of Ψa,b (t).
The time-frequency representation is performed by
repeatedly filtering the signal with a pair of filters that
cut the frequency domain in the middle. Specifically, the
discrete wavelet transform decomposes the signal into
approximation coefficients (CA) and detailed coeffi-
cients (CD). The approximation coefficient is subse-
quently divided into new approximation and detailed
coefficients. This process is carried out iteratively pro-
ducing a set of approximation coefficients and detail
coefficients at different levels or scales [23] (Table 1).
Table 1. Decomposition of EEG signals into different frequency
bands with a sampling frequency of 256 Hz.
tion Level Frequency
0–4 A5 Delta 4
4–8 D5 Theta 4
814 D4 Alpha 6
1432 D3 Beta 18
3264 D2 Gama 32
64–128 D1 Noises 64
In this work, the multi-resolution analysis of “db4”
wavelet function is used for decomposing the EEG sig-
nals into five levels and three frequency bands (alpha,
beta, and gamma) that are considered for deriving the
statistical features (Ta bl e 1 ). This wavelet function has
been chosen due to their near optimal time-frequency
localization properties. Therefore, extraction of EEG
signals features are more likely to be successful [24]. In
order to analyze the characteristic natures of different
EEG patterns, we proposed an energy based feature
called Recoursing Energy Efficiency (REE) in our earlier
work [15]. In that work, we use the Fuzzy C Means
(FCM) and Fuzzy K-Means (FKM) for clustering the
human emotions. The equation for deriving REE for
three frequency bands is given in Eq.4. In this present
work, we used the same feature and two of its modified
form namely Logarithmic REE (LREE) and Absolute
Logarithmic REE (ALREE) for classifying emotions
using two linear classifiers. The equations for calculating
LREE and ALREE for gamma band are given in Eq.5
and Eq .6 ; similarly the remaining frequency bands can
be derived. These features are derived from the three
frequency bands of EEG.
gamma b
total b
log gamma
gamma b
total b
log gamma
gamma b
total b
3total balphabetagamma
 (7)
In addition, we have used a set of conventional fea-
tures (power, standard deviation and variance) for clas-
sifying the emotions and comparing the efficacy of
newly proposed statistical features (Table 2).
2.5. Emotion Classification
In this work, we have used two simple classifiers
such as Linear Discriminant Analysis (LDA) and K
M. Murugappan et al. / J. Biomedical Science and Engineering 3 (2010) 390-396
Copyright © 2010 SciRes. JBiSE
Table 2. Statistical features (conventional) used for emotion
recognition and their description.
Features Description
Measures the deviations of electrodes potential
from the mean value
Power Measures the amplitude of EEG signal
Variance Measures the variability of electrodes potential
Nearest Neighbor (KNN) for classifying the discrete emo-
tions. Among these two classifiers, LDA provides ex-
tremely fast evaluations of unknown inputs performed
by distance calculations between a new sample and
mean of training data samples in each class weighed by
their covariance matrices. A linear discriminant analysis
tries to find an optimal hyper plane to separate five
classes of emotions (disgust, happy, surprise, fear and
In addition, KNN is also a simple and intuitive method of
classifier used by many researchers typically for classi-
fying the signals and images. This classifier makes a
decision on comparing a new labeled sample (testing
data) with the baseline data (training data). In general,
for a given unlabeled time series X, the KNN rule finds
the K “closest” (neighborhood) labeled time series in the
training data set and assigns X to the class that appears
most frequently in the neighborhood of k time series.
There are two main schemes or decision rules in KNN
algorithm, that is, similarity voting scheme and majority
voting scheme [25]. In our work, we used the majority
voting for classifying the unlabeled data. It means that, a
class (category) gets one vote, for each instance, of that
class in a set of K neighborhood samples. Then, the new
data sample is classified to the class with the highest
amount of votes. This majority voting is more com-
monly used because it is less sensitive to outliers. Be-
sides the training and testing samples, LDA does not
require any external parameter for classifying the dis-
crete emotions. However, in KNN, we need to specify
the value of “K” closest neighbor for emotions classifi-
cation. In this experiment, we try different “K” values
ranging from 2 to 6. Finally, the value of “K” is selected
as 5. This gives a maximum classification performance
among the other values of K.
Among all twenty subjects, we sample and preprocess a
total of 460 EEG epochs from five discrete emotions.
The number of data points in each epoch depends on the
time duration of video clips. In our experiment, the time
duration of video clips vary from one another. The next
stage is to train the KNN classifier with a best value of K
while LDA classifier directly works for classifying the
emotions. Among these two classifiers, LDA is a very
simple but elegant approach to classify various emotions.
The classification ability of a statistical feature set can
be measured through classification accuracy by averag-
ing five times over a 5 fold cross-validation. In order to
develop a reliable and efficient emotion recognition sys-
tem with less number of electrodes, we have compared
the classification accuracy of the original set of channels
(62) with reduced set of channels 24 channels [5] and 8
channels [21]. In [5], the reduced set of channels was
obtained on the basis of localizing the frequency range
of emotion over different areas of the brain though cog-
nitive analysis.
From Table 3, we inferred that, KNN gives higher
average classification accuracy than LDA on three dif-
ferent channels sets. The maximum classification accu-
racy of 83.26%, 79.93% and 72.68% is obtained using
ALREE feature on 62 channels (Figure 2(a)), 24 chan-
nels (Figure 2(b)) and 8 channels (Figure 2(c)) respec-
tively. Among the three different channel combination,
ALREE performs better than the other proposed (REE
and LREE) and conventional features. Ta ble 4 to Table 6
shows the individual emotions classification rate of
KNN and LDA classifiers for three different sets of
channels. From Ta bl e 4 , we found that, the 62 channel
EEG data gives the maximum individual classification
rate on five emotions (happy, surprise, fear and neutral)
compared to other channel sets. The maximum subsets
of emotions classification rate of 91.67% for disgust,
81.67% for happy and surprise, 81.25% for fear and
93.75% for neutral is achieved using 62 channels EEG
signals. Hence, the 62 channel EEG performs better over
other channels sets for classification of human emotion
through EEG.
Table 3. Average classification accuracy (%) of proposed and conventional statistical features using KNN and LDA.
62 Channels 24 Channels 8 Channels
Conventional Features
Power 55.87 63.40 64.71 45.43 70.51 45.22
Std Dev 62.46 71.73 70.58 54.78 72.61 53.26
Variance 56.10 64.56 68.84 46.67 71.52 44.06
Proposed Energy Features
REE 51.23 77.17 77.97 64.64 72.68 55.29
LREE 81.08 76.08 79.49 67.10 71.30 58.48
ALREE 83.26 75.21 79.93 66.30 72.68 58.12
M. Murugappan et al. / J. Biomedical Science and Engineering 3 (2010) 390-396
Copyright © 2010 SciRes. JBiSE
Table 4. Subsets of emotions classification rate (%) for 62 channel EEG over conventional and modified energy features.
Disgust Happy Surprise Fear Neutral
Conventional Features
Power 58.33 80 60 40 70 73.33 39.58 52.08 39.58 75
Std Dev 56.67 86.67 66.67 70 66.67 78.33 52.08 54.16 47.92 79.17
Variance 63.33 78.33 78.33 50 65 80 39.58 41.67 37.5 62.5
Proposed Energy Features
REE 53.33 90 60 73.33 63.33 78.33 22.92 60.42 39.58 93.75
LREE 91.67 91.33 78.33 58.33 73.33 73.33 68.75 75 89.58 91.92
ALREE 91.67 88.33 81.67 60 78.33
81.67 81.25 68.75 89.58 93.75
Table 5. Subsets of emotions classification rate (%) for 24 channel EEG over conventional and modified energy features.
Disgust Happy Surprise Fear Neutral
Conventional Features
Power 83.33 83.33 73.33 38.33 60 60 50 58.33 52.08 70.83
Std Dev 85 55 68.33 30 60 71.67 47.92 62.5 72.92 45.83
Variance 83.33 68.33 71.67 18.33 68.33 65 50 35.42 56.25 47.92
Proposed Energy Features
REE 91.67 90 65 28.33 70 76.67 60.42 45.83 72.92 75
LREE 90 91.67 71.67 36.67 76.67 73.33 66.47 56.25 73.75 77.08
ALREE 90 90 75 40 65
76.67 66.67 52.08 91.67 70.08
Table 6. Subsets of emotions classification rate (%) for 8 channel over conventional and modified energy features.
Disgust Happy Surprise Fear Neutral
Conventional Features
Power 90 48.33 70 40 71.67 83.33 62.5 16.67 60.42 45.83
Std Dev 86.67 70 71.67 26.68 58.33 70 64.58 39.58 77.08 39.58
Variance 83.33 50 63.33 35 61.67 86.67 62.5 14.58 77.08 20.83
Proposed Energy Features
REE 76.67 83.33 53.33 40 65 63.33 60.42 52.08 77.08 39.58
LREE 78.33 90 63.33 40 63.33 65 70.83 50 79.17 62.5
ALREE 80 85 61.67
38.33 70 55 75 52.08 81.25 70.08
In this paper we have presented an approach to discrete
emotion recognition based on the processing of EEG
signals. One of the major limitations on this research is
the lack of international standard data base. Hence, we
have compared the efficacy of emotion classification
using newly proposed energy features with conventional
features. Here, the modified energy features classify the
emotions better than the conventional features with
higher classification rate. The newly proposed feature,
Absolute Logarithmic Recoursing Energy Efficiency
gives the maximum average classification rate over other
conventional features. Therefore the extracted features
successfully capture the emotional changes of the sub-
ject through their EEG signals regardless of the user’s
cultural background, race, and age. In addition, it also
shows a significant relationship between EEG signals
and emotional states experienced by the subjects during
the interaction with audio-visual content. This study is
ongoing to involve different classification algorithms in
order to track the emotional status of brain activation
during audio-visual stimuli environment. The results of
this study provide a framework of methodology that can
be used to elucidate the dynamical mechanism of human
emotional changes underlying the brain structure. In
future, we are interested to develop efficient feature ex-
traction algorithm using different wavelet functions and
with a different set of statistical features for improving
M. Murugappan et al. / J. Biomedical Science and Engineering 3 (2010) 390-396
Copyright © 2010 SciRes. JBiSE
the emotion classification rate.
This work is supported by the two grants of Ministry of Science and
Technology Innovation (MOSTI), Malaysia. Grant Code: 9005–00020
& 9005–00002.
[1] Bung, H. and Furui, S. (2000) Automatic recognition and
understanding of spoken languages-a first step toward
natural human machine communication. Proceedings of
IEEE, 88, 1142-1165.
[2] Cowie, R., Douglas, E., Tsapatsoulis, N., Votsis, G., Kol-
lias, G., Fellenz, W. and Taylor, J.G. (2001) Emotion Rec-
ognition in human-computer interaction. IEEE Signal
Processing, 1, 3773-3776.
[3] Takahashi, K. (2004) Remarks on emotion recognition
from bio-potential signals. The Second International
Conference on Autonomous Robots and Agents, 186-191.
[4] Picard, R.W. (2000) Toward computers that recognize
and respond to user emotion. IBM Systems Journal, 39(3),
[5] Picard, R.W. and Healey, J. (1997) Affective wearable’s.
Personal Technologies, 1(4), 231-240.
[6] Kim, K.H., Band, S.W. and Kim, S.B. (2004) Emotion
recognition system using short-term monitoring of physio-
logical signals. Proceedings on Medical & Biological
Engineering & Computing, 42, 419-427.
[7] Chanel, G., Karim, A.A. and Pun, T. (2007) Valence-arousal
evaluation using physiological signals in an emotion re-
call paradigm. Lecturer Notes in Computer Science, 1,
[8] Chanel, G., Kronegg, J., Grandjean, D. and Pun, T. (2005)
Emotion assessment: Arousal evaluation using EEG’s
and peripheral physiological signals. Technical Report,
4105, 530-537.
[9] Murugappan, M., Rizon, M., Nagarajan, R. and Yaacob,
S. (2009a) Inferring of human emotion states using mul-
tichannel EEG. International Journal of Soft Computing
and Applications (IJSCA), EURO Journals, United King-
dom. (Accepted).
[10] Russell, J.A. (1980) A Circumplex model of affect. Jour-
nal of Personality and Social Psychology, 39(6), 1161-
[11] Lang. P.J. (1995) The emotion probe: Studies of motiva-
tion and attention. American Psychologist, 50(5), 372-
[12] Danny, O.B. (2008) Automated artifact detection in brain
stream. Technical Report, 1-8.
[13] Wang, Y. and Guan, L. (2005) Recognizing human emo-
tion from audiovisual information. Proceedings of the
International Conference on Acoustics, Speech, and Signal
Processing, 2, 1125-1128.
[14] Murugappan, M., Rizon, M., Nagarajan, R. and Yaacob,
S. (2008) Asymmetric ratio and FCM based salient chan-
nel selection for human emotion recognition using EEG.
WSEAS Transactions on Signal Processing, 10(4), 596-603.
[15] Murugappan, M., Rizon, M., Nagarajan, R., Yaacob, S.,
Hazry, D. and Zunaidi, I. (2008) Time-frequency analysis
of EEG signals for human emotion detection. IFMBE
Proceedings, 21, 262-265.
[16] Takahashi, K. and Tsukaguchi, A. (2004) Remarks on
emotion recognition from multi-modal bio-potential sig-
nals. Proceedings of IEEE International Workshop on
Robot and Human Interactive Communication, 95-100.
[17] Jonghwa, K. and Elisabeth, A. (2008) Emotion recogni-
tion based on physiological changes in music listening.
IEEE Transactions on Pattern Analysis and Machine
Intelligence, 30(12), 1-17.
[18] Murugappan, M., Rizon, M., Nagarajan, R. and Yaacob,
S. (2009) An Investigation on visual and audiovisual stimu-
lus based emotion recognition using EEG. Transactions
on Medical Engineering and Informatics, 1(3), 342-356.
[19] Jung, T. (2000) Removing electroencephalographic arti-
facts by blind source separation. Journal of Psychophysi-
ology, 37(2), 163-178.
[20] Gott, P.S., Hughes, E.C. and Whipple. (1984) Voluntary
control of two lateralized conscious states: Validation by
electrical and behavioral studies. Journal of Neuropsy-
chological, 22, 65-72.
[21] Mallat, S.G. (1989) A theory for multi-resolution signal
decomposition: The wavelet representation. IEEE Trans-
actions on Pattern Analysis and Machine Intelligence,
11(7), 674-693.
[22] Merzagora, A.C., Bunce, S., Izzetoglu, M. and Onaral, B.
(2006) Wavelet analysis for EEG feature extraction in
deceptive detection. IEEE Proceedings on EBMS, 6, 2434-
[23] Chethan, P. and Cox, M. (2002) Frequency characteris-
tics of wavelets. IEEE Transactions on Power Delivery,
17(3), 800-804.
[24] Charles, W.A. and Zlatko, S. (1997) Classification of
EEG signals from four subjects during five mental tasks.
IEEE Proceedings on Engineering Application in Neural
Networks, 407-414.
[25] Wanpracha, A.C., Ya-ju, F. and Rajesh, C.S. (2007) On
the time series k-nearest neighbor classification of ab-
normal brain activity. IEEE Transactions on Systems,
Man and Cybernetics–Part A: Systems and Humans,
37(6), 1005- 1016.