27-07-2012, 10:04 AM
Asymmetric Ratio and FCM based Salient Channel Selection for
Human Emotion Detection Using EEG
[attachment=29162]
Abstract:
Electroencephalogram (EEG) is one of the most reliable physiological signals used for detecting the
emotional states of human brain. We propose Asymmetric Ratio (AR) based channel selection for human emotion
recognition using EEG. Selection of channels reduces the feature size, computational load requirements and robustness
of emotions classification. We address this crisis using Asymmetric Variance Ratio (AVR) and Amplitude
Asymmetric Ratio (AAR) as new channel selection methods. Using these methods the 28 homogeneous pairs of EEG
channels is reduced to 4 and 2 channel pairs respectively. These methods significantly reduce the number of
homogeneous pair of channels to be used for emotion detection. This approach is illustrated with 5 distinct emotions
(disgust, happy, surprise, sad, and fear) on 63 channels EEG data recorded from 5 healthy subjects. In this study, we
used Multi-Resolution Analysis (MRA) based feature extraction the original and reduced set of channels for emotion
classification. These approaches were empirically evaluated by using a simple unsupervised classifier, Fuzzy C-Means
clustering with variable clusters.
Introduction
The electric potential measured at the scalp through a set
of electrodes (channels) are rich in information about the
brain activity. Researchers believe that the state of the
brain changes as emotions change, therefore the EEG is
suitable for the task of recording the changes in
brainwaves which vary in accordance with emotions [1].
Most of the useful information about the functional state
of human brain lies in the frequency range of DC-30 Hz.
EEG signals are discern as: delta band (0 to 4 Hz), theta
band (4 to 8 Hz), alpha band (8-12 Hz), beta band (12-
16 Hz), and gamma band (16- 30 Hz). The primary
research on emotion recognition depends on alpha band
[2]. In this work, alpha band of EEG frequency rhythm
is considered for both channel selection and feature
extraction.
Experimental Setup
Data Acquisition Protocol
In our work, we have used an audio-visual induction
based protocol for eliciting the primary emotions. To
elicit the target emotions in our study, we used two
commercial video clips out of 10 for each emotion. A
pilot panel study has been conducted on 10 subjects,
those who are not taking part in the experiment to select
any 2 video clips from the entire video clips sets. The
audio-visual stimulus protocol for two trials EEG
recording for our experiments is shown in Fig 2. In the
same way the other emotions over 5 subjects are
considered for acquiring the EEG signals. The x1 and x2
denote the time periods of selected video clips. Three
females and two males in the age group of 21-27 years
were employed as subjects in our experiment. Once the
consent forms were filled-up, the subjects were given a
simple introduction about the research work and stages
of experiment. The recording of EEG signal has been
done through Nervus EEG, USA with 64 channel
electrodes at a sampling frequency of 256 Hz and bandpass
filtered between 0.05 Hz and 70 Hz. The reference
electrode is placed between AF1 and AF2.
Feature Extraction
The multi-resolution analysis of wavelet transform is
used for decomposing the EEG signals into several
frequency bands. The statistical features from the EEG
signals for different emotions are derived from the
Alpha band. The basic methodology for feature
extraction using EEG signal on emotion recognition is
shown in Fig 4. 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 obtained by WT makes it as a good candidate
for the extraction of details as well as approximations of
the signal which cannot be obtained by Fast Fourier
Transform (FFT) or by Short Time Fourier Transform
(STFT) [11, 12].
Wavelet Transform
A wavelet is a small oscillatory wave which has its
energy concentrated in time. It has an ability to allow
simultaneous time and frequency analysis and it is a
suitable tool to analyze transient, non-stationary or timevarying
signals [13, 14]. 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.
Fuzzy C-Means (FCM) Clustering
In this experimental study, we used fuzzy c-means
(FCM) clustering [19,20] to assess the various emotions
from the overlapped electrophysiological signals.
Clustering is a tool that attempts to assess the
relationships among patterns of the data. The
organization of patterns within a cluster is more similar
to each other than patterns belonging to different
clusters. The FCM seeks to group the sampled data
together so as to minimize the variance (the objective
function) between data in the same cluster and maximize
the variance between data in different clusters. This
selection method allows each cluster of data to be
represented by a “cluster center”, where each center is a
representation of data geometrically closest to it. All
data sets can belong to all cluster centers, with a degree
of membership (DOM) in each cluster in the interval [0,
1]. The DOM is directly related to the Mahalanobis
distance between each data sample and the cluster center
[21].
Conclusions
Emotion occurs very differently based on the situation,
age, and growth environment etc. Even for a person, it
changes day in and day out. So, to get the statistical
output from several numbers of subject is difficult and
has no meaning [29]. In this work, we have performed
the experiment with 5 young and healthy subjects for
deriving salient EEG channels from the original set of
EEG channels for primary emotion detection. The
generalized solution for channel selection can be derived
by using a larger number of data sets. This paper has
proposed an approach to select subset of optimal
channels for emotion recognition using EEG signals.
Human Emotion Detection Using EEG
[attachment=29162]
Abstract:
Electroencephalogram (EEG) is one of the most reliable physiological signals used for detecting the
emotional states of human brain. We propose Asymmetric Ratio (AR) based channel selection for human emotion
recognition using EEG. Selection of channels reduces the feature size, computational load requirements and robustness
of emotions classification. We address this crisis using Asymmetric Variance Ratio (AVR) and Amplitude
Asymmetric Ratio (AAR) as new channel selection methods. Using these methods the 28 homogeneous pairs of EEG
channels is reduced to 4 and 2 channel pairs respectively. These methods significantly reduce the number of
homogeneous pair of channels to be used for emotion detection. This approach is illustrated with 5 distinct emotions
(disgust, happy, surprise, sad, and fear) on 63 channels EEG data recorded from 5 healthy subjects. In this study, we
used Multi-Resolution Analysis (MRA) based feature extraction the original and reduced set of channels for emotion
classification. These approaches were empirically evaluated by using a simple unsupervised classifier, Fuzzy C-Means
clustering with variable clusters.
Introduction
The electric potential measured at the scalp through a set
of electrodes (channels) are rich in information about the
brain activity. Researchers believe that the state of the
brain changes as emotions change, therefore the EEG is
suitable for the task of recording the changes in
brainwaves which vary in accordance with emotions [1].
Most of the useful information about the functional state
of human brain lies in the frequency range of DC-30 Hz.
EEG signals are discern as: delta band (0 to 4 Hz), theta
band (4 to 8 Hz), alpha band (8-12 Hz), beta band (12-
16 Hz), and gamma band (16- 30 Hz). The primary
research on emotion recognition depends on alpha band
[2]. In this work, alpha band of EEG frequency rhythm
is considered for both channel selection and feature
extraction.
Experimental Setup
Data Acquisition Protocol
In our work, we have used an audio-visual induction
based protocol for eliciting the primary emotions. To
elicit the target emotions in our study, we used two
commercial video clips out of 10 for each emotion. A
pilot panel study has been conducted on 10 subjects,
those who are not taking part in the experiment to select
any 2 video clips from the entire video clips sets. The
audio-visual stimulus protocol for two trials EEG
recording for our experiments is shown in Fig 2. In the
same way the other emotions over 5 subjects are
considered for acquiring the EEG signals. The x1 and x2
denote the time periods of selected video clips. Three
females and two males in the age group of 21-27 years
were employed as subjects in our experiment. Once the
consent forms were filled-up, the subjects were given a
simple introduction about the research work and stages
of experiment. The recording of EEG signal has been
done through Nervus EEG, USA with 64 channel
electrodes at a sampling frequency of 256 Hz and bandpass
filtered between 0.05 Hz and 70 Hz. The reference
electrode is placed between AF1 and AF2.
Feature Extraction
The multi-resolution analysis of wavelet transform is
used for decomposing the EEG signals into several
frequency bands. The statistical features from the EEG
signals for different emotions are derived from the
Alpha band. The basic methodology for feature
extraction using EEG signal on emotion recognition is
shown in Fig 4. 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 obtained by WT makes it as a good candidate
for the extraction of details as well as approximations of
the signal which cannot be obtained by Fast Fourier
Transform (FFT) or by Short Time Fourier Transform
(STFT) [11, 12].
Wavelet Transform
A wavelet is a small oscillatory wave which has its
energy concentrated in time. It has an ability to allow
simultaneous time and frequency analysis and it is a
suitable tool to analyze transient, non-stationary or timevarying
signals [13, 14]. 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.
Fuzzy C-Means (FCM) Clustering
In this experimental study, we used fuzzy c-means
(FCM) clustering [19,20] to assess the various emotions
from the overlapped electrophysiological signals.
Clustering is a tool that attempts to assess the
relationships among patterns of the data. The
organization of patterns within a cluster is more similar
to each other than patterns belonging to different
clusters. The FCM seeks to group the sampled data
together so as to minimize the variance (the objective
function) between data in the same cluster and maximize
the variance between data in different clusters. This
selection method allows each cluster of data to be
represented by a “cluster center”, where each center is a
representation of data geometrically closest to it. All
data sets can belong to all cluster centers, with a degree
of membership (DOM) in each cluster in the interval [0,
1]. The DOM is directly related to the Mahalanobis
distance between each data sample and the cluster center
[21].
Conclusions
Emotion occurs very differently based on the situation,
age, and growth environment etc. Even for a person, it
changes day in and day out. So, to get the statistical
output from several numbers of subject is difficult and
has no meaning [29]. In this work, we have performed
the experiment with 5 young and healthy subjects for
deriving salient EEG channels from the original set of
EEG channels for primary emotion detection. The
generalized solution for channel selection can be derived
by using a larger number of data sets. This paper has
proposed an approach to select subset of optimal
channels for emotion recognition using EEG signals.