Weitere ähnliche Inhalte
Ähnlich wie Lvq based person identification system
Ähnlich wie Lvq based person identification system (20)
Mehr von IAEME Publication
Mehr von IAEME Publication (20)
Kürzlich hochgeladen (20)
Lvq based person identification system
- 1. International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 4, July-August (2013), © IAEME
185
LVQ BASED PERSON IDENTIFICATION SYSTEM
Nisargkumar Patel1
, Prof. V.V. Shete2
, Ashwini Charantimath3
MIT College of Engineering, Pune, India
ABSTRACT
A wide verity of security system requires reliable recognizable scheme to either confirm or
identify the person from a group. Recently, researchers taking great interest in physiological like
EEG and ECG. It is also known as EEG or ECG signature of a person. Due to characteristics of EEG
signal, it is very hard to breach the system or copy the EEG signal. In this paper EEG based method
is introduced for person identification. There are different methods which can be used for feature
extraction. Here, in this paper Wavelet Packet Decomposition (WPD) is used. For classification
Learning Vector Quantization (LVQ) is used.
Keywords: EEG signal processing, Bio-metric security, person identification, Learning Vector
Quantization (LVQ).
INTRODUCTION
The term “Biometric” is known as a system which uses human biometric characteristics for
identification of a person. Body characteristics like face and voice are used to recognize each other
from thousands of years [2]. Recently, researchers are focused on the biometrics system for prevent
precious data or transaction from forgery.
In conventional biometrics security systems are based on the biometric traits such as 1)
Figure print. 2) Retinal or iris Scanning 3) DNA. 4) Voice reorganization etc [6]. Identification and
authentication are two different processes, for authentication of a person it is necessary to identify
the person. Identification of a person can be defines as it is the process to identify individual from a
group and authentication is defines as it is the process of confirm or deny the identity claim of a
person. In this method we can use any human characteristic and or physiological signal which should
satisfy characteristics such as Universality, Distinctiveness, Permanence, Collectability,
Acceptability, and Circumvention.
The point is most of the conventional traits does not satisfy above characteristics, for example
figure print and voice reorganization are not fulfil above characteristics because accidently or
handicap person may not have figure print or voice so it is not universal. Solution for above problem
INTERNATIONAL JOURNAL OF COMPUTER ENGINEERING &
TECHNOLOGY (IJCET)
ISSN 0976 – 6367(Print)
ISSN 0976 – 6375(Online)
Volume 4, Issue 4, July-August (2013), pp. 185-193
© IAEME: www.iaeme.com/ijcet.asp
Journal Impact Factor (2013): 6.1302 (Calculated by GISI)
www.jifactor.com
IJCET
© I A E M E
- 2. International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 4, July-August (2013), © IAEME
186
is that, we have to concentrate on the more unique characteristics of a human like ECG and EEG
signals of a human. This is a unique pattern of any human in his entire life. EEG signal is the
electrical signal generated by the brain during its activity and it is recorded on the scalp of the
human. Here in this paper we are using an EEG pattern for the identification of a human or person,
because it is different in every subject. In other words inter-subject variability is very high with EEG
signal.
This method for identification of a human uses EEG signal as human traits. Feature
extraction of the EEG signal can b done with various methods deepens upon the application, how
much accuracy is required. In this method FFT and Wavelet packet decomposition is used. Also for
classification we can use both supervised and unsupervised learning methods and algorithms for
artificial neural network. Here, learning vector quantization is used for the classification of signals.
Se´bastien Marcel [2], was investigated the use of brain activity for person authentication. It
has been shown in previous studies that the brain-wave pattern of every individual is unique and that
the electroencephalogram (EEG) can be used for biometric identification. Jiang Feng Hu [16] was
able to achieve accuracy ranging from 75% for subject authentication and 75% to 78.3% for
identification using 6 channels. Learning Vector Quantization (LVQ) neural network was used for
the classification that achieved classification scores in the range of 80% to 100%, (case dependent).
METHODOLOGY OF SYSTEM
The main intention of this paper is to introduce a one of the best method for human
identification for improve the security systems. Since everybody have unique EEG pattern which is
universal and brain damage is rarely occurred. In this system there are some steps for getting
accurate result or we can say some methodology.
Figure 1: Block Diagram of Proposed System
Firstly, we have to acquire a signal from the subject it is the EEG signature of the subject
then find out which feature suits for the application and find out the method for extraction of the
same from the signal and then it will be given to the classifier which classifies it in the different class
depends upon its algorithm. Selection of a classifier is also essential and it also depends upon
application as well. All the steep discussed above will be explained in this section and it is illustrated
in figure 1.
1. Data Acquisition
Signal or data acquisition is the most important step because other steps are depending on the
data. EEG data is taken from the scalp of the subject through the electrode (sensors).
Superconductive gel is place in between the sensor and the scalp because signal coming from the
brain through scalp is very weak. It is in terms of micro volt. Thus, better reception of the signal
conductive gel is used. And signal detected from the particular node is not the signal from that
EEG
signal
Pre-
processing
Security
check
EEG
feature
extraction
EEG pattern
classification
- 3. International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 4, July-August (2013), © IAEME
187
particular place where the node is placed but it is the average of the total relative potential of the
scalp.
Figure 2: Standard 10-20 international system for sensor placement [13]
Raw EEG data is too noisy so it needs to be filtered and amplified for further processing.
Subject is requiring sitting at rest with eye closed in silent room this condition makes easy to yield
required rhythm.
In 10-20 international system sensor is place such that distance between two sensors is either
10% or 20% of total area of the scalp. For that measurement of scalp is taken before node placement,
and then it is divided in two equal spaces which contain 10% or 20% distances from each node.
2. Signal Pre-processing
At the time of data acquisition signal amplification and filtration process is applied to that
data. But additional digital band pass filter is applied in second stage because we need specific band
of the signal EEG signal band is 0.5 to around 90 Hz. For our application we require signal of 0.5 to
64 Hz. To acquire the same band Butterworth band pass filter is applied and due to this operation
some line interference and noise is also removed from the signal. EEG signal dived in four basic
EEG frequency patterns which define the mental condition of a person like person is in drowsy mood
or he/she is fully awake state, all patter is being described below [6]:
• Delta ( ) (0.5 - 4Hz): These waves are primarily associated with deep sleep and may be with
waking state.
• Theta (θ) (4 - 8 Hz): These waves have been associated with access to unconsciousness creative
inspiration and deep meditation it seems to be related to the level of arousal.
• Alpha ( ) (8 - 13 Hz): These waves have been associated with a relaxed awareness without
attention or concentration.
• Beta (β) (>13 Hz): It is most evident in frontal region and associative with busy or anxious
thinking and active concentration.
After getting desired pattern of the EEG signal next task is to extract the features from the
desired EEG pattern (8-12 Hz) with different methods.
3. Feature extraction
Feature extraction is the process to separate desired output from the EEG pattern. Rather than
apply the operation on EEG pattern the reason behind the feature extraction is classification process
is easy with the extracted feature because, with help of feature we can easily differentiate the person.
There are various methods are available for feature extraction [9] [11] [13]. Few techniques are listed
below:
• ARR model,
• Fast Fourier Transform (FFT),
- 4. International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 4, July-August (2013), © IAEME
188
• Discreet Fourier Transforms (DFT),
• Wavelet packet decomposition (WPC)
From above techniques we can use any of it according to application. Here in this paper WPC
is used for the feature extraction.
3.1.Wavelet Packet decomposition (WPD)
To overcome the disadvantage of the FFT here WPD is used for feature extraction because it
is able to give the solution in time as well as in frequency domain of the non-stationary signal as
well. In the discreet wavelet transform, each level is calculated by passing only previous
approximation coefficients through low and high pass quadrature. WPD is the wavelet transform
where the signal is passed through more filter than the discrete wavelet transform (DWT) as shown
in figure 3 [1]. For n levels of decomposition the WPD produces 2n different sets of coefficients (or
nodes) as opposed to (3n + 1) sets for the DWT. Studies using wavelet packet decomposition to
analyze EEG signals were able to obtain the four brain rhythms: alpha, beta, theta and delta.
Figure 3: Wavelet Packet Decomposition over 2 levels
The signal x(t) is decomposed into different scales in equation (1) [3]
ݔሺݐሻ ൌ ݀ሺ݇ሻ ߮,ሺݐሻ ܽሺ݇ሻ ,ሺݐሻ
∞
ୀି∞
∞
ୀି∞
ୀଵ
. . ሺ1ሻ
Where j is the scale parameter, ߮,ሺݐሻ are discrete analysis wavelets and ,ሺݐሻ are
discrete scaling functions. ݀ሺ݇ሻ Are the wavelet coefficients at scale 2
and ܽሺ݇ሻ are the scaling
coefficients at scale 2K
.
In this research four level of wavelet packet decomposition is use for selection of alpha band of
the EEG signal which is collected from (4, 2) node of wavelet packet tree. Detail co-efficient of the
signal of 8 to 12Hz are extracted for further processing. For the classification we can use different
features that are listed below [1] [9] [10]:
• Mean of the signal.
• Standard deviation.
• Entropy.
• Minimum value of the signal.
• Maximum value of the signal.
X[n]
H1
H1H2
H1L2
L1
L1H1
L1L2
- 5. International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 4, July-August (2013), © IAEME
189
Here in this paper detail coefficient of the alpha signal which are lies between 8 to 12 Hz are
extracted. Here minimum, maximum, mean, and standard deviation these features are used for the
further process. Value of each coefficient is calculated using following equations [1]:
µ௫
ൌ
1
݊
ݔ
ୀଵ
… … … … … … … … … . ሺ2ሻ
ߪ௫ ൌ ඩ
1
݊
ሺݔ െ ߤ௫ሻଶ
ୀଵ
… … … … … ሺ3ሻ
ߝሺݔሻ ൌ െ ݔଶ
ሺݐሻ log ݔଶ
ሺݐሻ
௧
… … … … . ሺ4ሻ
Thus total five feature of the one channel such four channels are used in this project so total
number of feature for each data is 5*4=20 feature for each person. Lastly, all the feature of all data is
store in one metrics which is used for training for the neural network.
4. Classification
Many researcher use artificial neural network for the classification purposes in many areas.
Classification of any signal or input is to verify or categorize in the pre-define classes. We can also
say classification is the process to differentiate the input in different range. Now a day it is very
popular because of its speed of calculation and data handling capacity is very high.
Figure 4: Architecture of LVQ neural network
In this paper learning vector quantization neural network is used learning vector quantization
is a standard statistical clustering technique for classifies the input to the specified output classes
after training LVQ assign weight vector to the input pattern closest to the output unit. And for
training “learnlvq1” algorithm is used for the classification. Here data is given to the network for
training thus, it is supervised learning technique we already know the output of the system. Basically,
network output is the signal is lying in which predefined classes.
Architecture of the LVQ neural network is shown below in figure 4. In this network there are
basically three layer, first is input layer in that number of neuron is depends upon the number of
inputs then hidden layers it can be one or more according the necessity of the application then lastly,
output layer.
- 6. International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 4, July-August (2013), © IAEME
190
The classifier performs a series of operation with each pattern of the training set until the
stopping criteria are met. When classification is done using LVQ desired class is given to the output
neurons in output layer. There are no certain rules for choosing number of hidden layer neurons.
Many tests are performing for optimum solution for the neural network configuration that will be
explained in the experimental results.
Figure 5: Flow Chart of LVQ Algorithm
EXPERIMENTAL RESULTS
Experiment is done with four sets of data and combination of subject is selected randomly.
During the process of recording the data, subject should sit at rest with eye close in a silent room and
remain calm throughout the whole process. In this experiment data is taken from channel F, C, P and
O at sampling frequency 240 Hz for 3 second epoch. Here results are shown below:
1. Data of EEG signal is lying between 0.5 to about 50Hz. Here for the further process EEG signal
of 0.5 to 50Hz is passing through butter-worth band pass filter which has pass-band frequency is
8Hz and stop-band frequency is 12Hz (alpha rhythm). Thus we get the alpha rhythm of subject’s
signal. All three channels follow the same operation. Following figure 6 shows the filtering
operation and resulting alpha rhythm of channel F3.
START
STOP
INITIALIZE WEIGHT AND
LEARNING RATE
REDUCE THE LEARNING RATE
CALCULATE EUCLIDEAN
DISTANCE AND UPDATE WEIGHT
INPUT VECTOR
IF DESIRED
OUTPUT?
IF DESIRED
OUTPUT?
UPDATE WEIGHT VECTOR
YES
YES
NO
NO
- 7. International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 4, July-August (2013), © IAEME
191
Original signal of channel F3
Filtered signal of channel F3
Figure 6: Original signal and filter signal of channel F3
2. Four level of wavelet packet decomposition is applied in a filtered signal. Here detail coefficient
of the (4, 2) component is reconstructed and used to extract the feature from every channel.
Features that are used in this method are: mean minimum, maximum, standard deviation and
entropy of the signal. All features are calculated with equation which is mention in above
explanation. These feature are stored in a metrics which will be used for the training of a neural
network, it is shows below in table 1.
Table 1: feature set of Data 1
Feature set of Data 1
F3 C3 P3 O1
-2.656x103
-2.764x103
-3.283x103
-3.24x103
2.851x103
2.679x103
3.253x103
3.749x103
-3.319 -2.457 -2.734 -1.220
42.146 30.685 33.849 14.556
3.128x108
4.668x108
7.752x108
7.976x108
Learning vector quantization neural network is train by the “learnlv1” algorithm is use for the
classification in this paper. The number of neuron in the input layer varied accordingly to the number
of the feature is used here in this experiment 5 feature of each channel is used for the training. Output
layer contains neuron according to the data base in other words it is depends on the class defined by
the user. Fir each feature extraction method, the configuration that produces weight arte optimized to
maximum classification, sensitivity and accuracy. Thus we can say that configuration of the
classifier plays important role in the experiments to achieve best classification.
Sensitivity ൌ
TP
ሺTP FNሻ
… … … … … . ሺ5ሻ
Speciϐicity ൌ
TN
ሺTN FPሻ
… … … … … . ሺ6ሻ
Accuracy ൌ
TP TN
ሺTP FP TN FNሻ
… … ሺ7ሻ
Where,
TP= True positive; TN=True negative;
FP=False positive; FN=False negative.
- 8. International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 4, July-August (2013), © IAEME
192
Figure 7: ROC of the LVQ neural network
Table 2: Performance of the LVQ Neural network
Parameters
Data set
1
Data set
2
Data set
3
Data set
4
TP 20 20 20 20
TN 04 04 04 04
FP 00 02 03 01
FN 00 02 03 01
Sensitivity
(%)
100.00 90.91 86.96 95.24
Specificity
(%)
100.00 66.67 57.14 80.00
Accuracy (%) 100.00 85.71 80.00 92.31
CONCLUSION AND FUTURE WORK
This experiment is implemented in Matlab 2011 and using neural network tool box. Network
sensitivity, specificity and accuracy is calculated using equation using following equation and value
of the each element in the table 2 and ROC of the network is also shown in figure 7.
This paper proposes one of the best and unique methods for biometric security system. Role
of the feature extraction from the signal is explained very well in this paper. Technique for the
feature extraction is very crucial for any kind of system, because classification process is only based
on that feature which is extracted from the signal. Here, Wavelet packed decomposition technique is
used for it and learning vector quantization neural network performs good and give 80 to 100%
accuracy for all four data set of the different subject. In future system testing for more data and
maintain and improve accuracy to 100% for more number of data. Here, four channels are used for
identification process; in future research can be done on how we can reduce the number of the
channel for this application with same accuracy and sensitivity. It is the main part of the future work.
- 9. International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 4, July-August (2013), © IAEME
193
REFERENCE
[1] Howida AbdelFattah Shedeed Faculty of Computers and information Sciences Ain Shams
University “New Method for Person identification in a Biometric Security System Based on
Brain EEG Signal Processing” 2011 IEEE.
[2] Se´bastien Marcel and Jose´ del R. Milla´n “Person Authentication Using Brainwaves (EEG) and
Maximum A Posteriori Model Adaptation” IEEE Transaction on pattern analysis and machine
intelligence vol. 29, No. 4, April 2007.
[3] S. Prabhakar, S. Pankanti, and A. K. Jain, “Biometric recognition: Security and privacy
concerns,” IEEE Security Privacy Magazine, vol. 1, no. 2, pp. 33–42, 2003
[4] K. Ravi and R. Palaniappan. “Leave-one-out authentication of persons using 40Hz EEG
oscillations.” In The International Conference on Computer as a Tool, volume 2, Belgrade, Serbia
& Montenegro, pages 1386 –1389, 2005.
[5] Wu Ting, Yan Guo-zheng, Yang Bang-hua, Sun Hong “EEG feature extraction based on wavelet
packet decomposition for brain computer interface”, 618–625 Elsevier Ltd 2008.
[6] Anil K. Jain, Arun Ross and Salil Prabhakar “An Introduction to Biometric Recognition” IEEE
Transaction on circuit and systems for video technology vol. 14, no. 1 4-20 January 2004.
[7] R. Brunelli and D. Falavigna, “Person identification using multiple cues,” IEEE Trans. Pattern
Anal. Machine Intelligence. vol. 12, pp. 955–966, Oct. 1995.
[8] Ramaswamy Palaniappan Danilo P. Mandic “EEG Based Biometric Framework for Automatic
Identity Verification” Journal of VLSI Signal Processing 49, 243–250, 2007.
[9] Carlos Guerrero-Mosquera and Angel Navia Vazquez“New approach in features extraction for
EEG signal detection” 31st
Annual International Conference of the IEEE EMBS Minneapolis,
Minnesota, USA, September 2-6, 2009.
[10] Fei Su, Liwen Xia, Anni Cai, Yibing Wu , Junshui Ma “EEG-based Personal Identification: from
Proof-of-Concept to A Practical System”, International Conference on Pattern Recognition 1051-
4651/10 © 2010 IEEE.
[11] Gelareh Mohammadi, Parisa Shoushtari, Behnam Molaee Ardekani and Mohammad B.
Shamsollahi “Person Identification by Using AR Model for EEG Signals”, Pwaset Volume 11
page 281-285 FEBRUARY 2006.
[13] Abdul-bary Raouf Suleman, Toka-Hameed Fatehi “Feature Extraction techniques of EEG signal
for BCI application”, 2007.
[14] A. Anoklin, O. Fisher, Y. Mao, P. Vogt, E. Schalt, F. Vogel,(1999), “A genetic study of the
human low-voltage electroencephalogram” Human Genetic ,vol. 90,PP. 99-112,1992.
[15] P. P. Paul and M. M. Monwar, “Human iris recognition for biometric identification” Proceedings
of 10th International Conference on Computer and Information Technology, 2007, pp. 44-48.
[16] H. Jian-Feng," Multi-feature biometric system based on EEG signals", In Proceedings of the 2nd
International Conference on Interaction Sciences, Seoul, Korea, 2009, pages 1341–1345.
[17] S. N. Sivanandam , S sumathi and S. N deepa “Introduction to neural networks using MATLAB
6.0”, The Mcgraw-Hill 2010.
[18] Donna L. Hudson and Maurice E. Cohen “Neural network and artificial intelligence for
biomedical engineering”, IEEE press 2000.
[19] Imteyaz Ahmad, F Ansari and U.K. Dey, “A Review of EEG Recording Techniques”,
International Journal of Electronics and Communication Engineering & Technology (IJECET),
Volume 3, Issue 3, 2012, pp. 177 - 186, ISSN Print: 0976- 6464, ISSN Online: 0976 –6472.
[20] Sayeesh and Dr. Nagaratna P. Hegde, “A Comparison of Multiple Wavelet Algorithms for Iris
Recognition”, International Journal of Computer Engineering & Technology (IJCET), Volume 4,
Issue 2, 2013, pp. 386 - 395, ISSN Print: 0976 – 6367, ISSN Online: 0976 – 6375.
[21] V.V.Ramalingam, S.Ganesh kumar and V. Sugumaran, “Analysis of EEG Signals using Data
Mining Approach”, International Journal of Computer Engineering & Technology (IJCET),
Volume 3, Issue 1, 2012, pp. 206 - 212, ISSN Print: 0976 – 6367, ISSN Online: 0976 – 6375.