View Full Paper

USING EIGENFACE FEATURES TO DRIVE SUPPORT VECTOR
MACHINES IN FACE RECOGNITION SYSTEMS
1
SHEELA SHANKAR, 2V.R UDUPI
1
Department of Electronics & Communication Engg, KLE Dr. M. S. Sheshgiri CET, Udyambag, Belgaum, India
2
Department of Electronics and Communication Engg, Gogte Institute of Technology, Belgaum, India
Abstract- Face recognition is one of the growing fields of research which has deep rooted applications in authentication
domains. Its success can be attributed to the various algorithms which have strived to make it work in the real time. The
major factor underlying such systems is the proper feature extraction and effective classification. The paper aims at using the
two most well-known techniques to accomplish the above mentioned factors. They are Eigenface method -to extract features
and Support vector Machines (SVM) –to classify the data. It was found that the proposed methodology was effective in
terms of classification and due to the ease of implementation; it can be adopted in real-world applications.
Keywords- Authentication, Face recognition, Eigenface Method, SVM.
mouth, etc. These are known as principal components
or eigenfaces. Eigen faces are the characteristic
features of an image. Principal Component Analysis
(PCA) is often used to extract the above discussed
patterns. PCA transforms an image into a set of
eigenfaces. The original image can be reconstructed
from the obtained eigenfaces. Each eigenface is
symbolic of only a specific characteristic of the face.
Reconstruction of the original image using these
eigenfaces basically involves the building of
weighted sum of all eigenfaces. The degree up to
which the specific characteristic is prevalent in the
original image is a function of the weight. An
enhanced version of eigenface approach was used by
minimizing the intrasubject variation and maximizing
the intersubject variation in.
I. INTRODUCTION
Face recognition is a form of identifying a given face
image and matching it against a set of faces in the
database, in order to authenticate or validate a person
under test. Today there are a number of face
recognition algorithms which have made this
technique feasible and attainable. Hence the method
has proved to be the most secure and efficient among
various biometric based authentication techniques.
Even then, face recognition has been an active
research domain even after decades of its
introduction. This is due to the factors that have
curbed its ease of implementation in real time
applications. The major reason being the variation in
faces due to facial hair, illumination, age, change in
poses, goggles and other accessories, etc.
III. SUPPORT VECTOR MACHINES
To overcome these challenges, more stress is laid
currently on effective extraction of features from the
given face and accurate classification of the datasets
to uniquely identify a person. The proposed study
uses Eigenface method to achieve the former and
Support Vector Machine (SVM) based classification
methodology to accomplish the latter. The rest of the
paper is organized as follows. Section II deals with a
detailed review of Eigenface method and Section III
deals with SVM. Section IV elaborates the
implementation of the proposed methodology. The
results are given in Section V. The paper concludes in
Section VI.
Support Vector Machine (SVM) is associated with
machine learning domain. The credit of formulating it
goes to Vladimir N. Vapnik. They are basically
supervised learning models to scrutinize the data and
identify patterns.
This is basically used for regression analysis and
classification. More specifically, a SVM builds a
boundary or classifier to distinguish or classify a set
of data. It is a straight line in case of 2 D feature
vector, while in 3D, it is a plane. For other higher
dimensions, it is called hyperplanes.
A good classifier is supposed to provide a larger
separation between the feature vectors of different
classes. Intuitively, this is to assure that the
classification is general when the feature vectors are
subjected to change due to noise.
Many of the traditional algorithms for face
recognition are based on an assumption that the
classifier once trained, cannot be subjected to
amendments during run-time. Hence it requires the
II. EIGENFACE METHOD
Face recognition is the technique of identifying a
specific fixed face amidst an assortment of faces. The
input signals to such model are images which are
subjected to variations like occlusion, illumination,
age, etc. In spite of these inhibitions, there exist
patterns which aid the purpose of face recognition.
The patterns here are the eyes, nose, skin colour,
Proceedings of 10th IRF International Conference, 04th October-2014, Bengaluru, India, ISBN: 978-93-84209-56-8
55
Using Eigenface Features to Drive Support Vector Machines in Face Recognition Systems
entire data specific to a face to be present during
training period, which is highly not feasible in real time applications. However, SVM classifiers
overcome this shortcoming by identifying the faces
that are not present in the database. These unknown
faces can be used to retrain the classifier.
Let M be the margin width, then,
SVM algorithm is extensively used in applications
like handwritten character recognition, text and
hypertext categorization, Bioinformatics (Protein
classification, Cancer classification), brain-wave data
analysis and image classification. Face images with
occlusions can also be recognized by SVM.
IV. IMPLEMENTATION
Face images for the study were taken from the AR
Face database. The details of the face database used
in the study are as given below:
A. AR Face Database
The database was created by Aleix Martinez and
Robert Benavente in the Computer Vision Center
(CVC) at the U.A.B. in the year 1988. It was the first
database to include occlusions. The total number of
subjects used here is 126 with 4000 total images.
Fig.1. shows a linear SVM classifier for two classes
of samples. Note that the points lying on the lines
wx+b =+1 and wx+b = -1 are called support vectors.
Similarly the method can be extended for multi class
problems.
It provides provision for variation in illumination,
frontal poses, expression, scarves, eye glasses, etc.
The size of the RGB colour images is 768 × 576
pixels. In a 2-week interval, the subjects face images
were captured twice by subjecting them to 13
different conditions.
The method deals with drawing a boundary line to
separate the vectors based on their types. Such a
boundary is called hyperplane. In case of n
dimensional problem, the dimension of the
hyperplane is n-1. Vectors that are closest to the
hyperplane are support vectors. The distance between
the support vectors is called the Margin M.
Fig. 2a shows samples taken from the training
database and Fig. 2b shows samples taken from the
testing database.
Fig.2. Images in the training database
Fig. 2b Images in the testing database.
Proceedings of 10th IRF International Conference, 04th October-2014, Bengaluru, India, ISBN: 978-93-84209-56-8
56
Using Eigenface Features to Drive Support Vector Machines in Face Recognition Systems
Fig. 3. Overall system implementation
The overall system implementation is as shown in
Fig. 3. Eigenfaces are derived for both the images in
the training and the testing databases. Fig 4a, b shows
the original and the derived eigenface, respectively.
Then their corresponding weights are calculated.
Their mean, standard deviation and kurtosis values
were found. This was necessary in order to coalesce
the matrix of eigen weights into a single numeric
value.
Fig. 5 and 6 shows the snapshots of the interface of
the developed package. Provisions like feature
extraction, SVM testing, extracting features from all
the faces at once, resetting of the transactions and
exiting the program are provided. Fig. 5 depicts the
extraction of features from the selected figure. The
figure under processing is shown in the space at the
right of the interface. Fig. 6 shows action performed
on using the SVM classification event. The face in
the training database with the nearest match is shown
here. This was continued in three iterations, one each
using mean, standard deviation and kurtosis. It was
found that the experimentation carried out using
kurtosis outperformed the rest.
The kurtosis of a distribution is defined as
k=
(
µ)
where µ is the mean of x, σ is the standard deviation
of x, and E(t) represents the expected value of the
quantity t.
These are then fed individually as inputs to the SVM
classifier. Totally, the classification was done for
three times, based on the results fed to the classifier.
Since face images of 10 different people were used, a
multi SVM was implemented in Matlab.
Fig. 4a.Original Image
V. RESULTS AND DISCUSSIONS
Fig. 4a shows the original image and its
corresponding Eigenface is shown in Fig. 4b.
Eigenface helps in better acquisition of features from
the original images which are ready to be used for
further processing.
Fig. 4b.Eigenface of the original image
Proceedings of 10th IRF International Conference, 04th October-2014, Bengaluru, India, ISBN: 978-93-84209-56-8
57
Using Eigenface Features to Drive Support Vector Machines in Face Recognition Systems
[3]
Sheela Shankar, Dr. V. R Udupi, “A Review on the
Challenges Encountered in Biometric Based Authentication
Techniques”, Volume 2, Issue 6, June 2014, International
Journal of Advance Research in Computer Science and
Management Studies (IJARCSMS). ISSN: 2321-7782
(Online).
[4]
Turk, M.A., Pentland, A.P., 1991. Eigenfaces
recognition. J. Cognitive Neurosci. 3 (1), 71–86.
[5]
Perlibakas, V., 2004. Distance measures for PCA-based face
recognition. Pattern Recognition Lett. 25 (6), 711–724.
[6]
Yang, J., Zhang, D., Frangi, A.F., Yang, J.Y., 2004.
Twodimensional PCA: A new approach to appearancebased face representation and recognition. IEEE Trans.
Pattern Anal Machine Intell. 26 (1), 131–137.
[7]
Jie Wang, K.N. Plataniotis, A.N. Venetsanopoulos,
“Selecting discriminant eigenfaces for face recognition”,
Pattern Recognition Letters 26 (2005) 1470–1482,
doi:10.1016/j.patrec.2004.11.029
[8]
W. –S Kang and J.Y. Choi, Kernel machine for fast and
incremental learning of face. Proc. of the International Joint
Conference SICE-ICASE. pp. 1015-1019. 2006
[9]
V. Vapnik, Statistical Learning Theory, Wiley, New York,
1998.
Fig. 5 Snapshot of the interface showing the feature extraction
event.
for
[10] Lin, Yuan-Pin, Chi-Hong Wang, Tien-Lin Wu, Shyh-Kang
Jeng, and Jyh-Horng Chen. "Support vector machine for
EEG signal classification during listening to emotional
music." In Multimedia Signal Processing, 2008 IEEE 10th
Workshop
on,
pp.
127-130.IEEE,
2008.
Doi:
10.1109/MMSP.2008.4665061
Fig. 6.Snapshot of the interface showing the SVM results.
[11] Panda, R., P. S. Khobragade, P. D. Jambhule, S. N. Jengthe,
P. R. Pal, and T. K. Gandhi. "Classification of EEG signal
using wavelet transform and support vector machine for
epileptic seizure diction." In Systems in Medicine and
Biology (ICSMB), 2010 International Conference on, pp.
405-408.IEEE, 2010. Doi:10.1109/ICSMB.2010.5735413
CONCLUSION
The paper deals with using the two robust algorithms
to achieve better recognition rates. Eigenface
approach was used to extract the features from the
face images which were then fed to the SVM
classifier for classification. A package was developed
in Matlab using these algorithms. The method was
more accurate when kurtosis method was used to
condense the eigen weights before using the SVM
classifier. It was found that the proposed method was
effective in terms of classification and thus fosters the
use of face recognition in real world domains.
[12] Kazuhiro Hotta, “Robust face recognition under partial
occlusion based on support vectormachine with local
Gaussian summation kernel”, Image and Vision Computing
26 (2008) 1490–1498. Elsevier. doi:10.1016/ j.imavis.
2008.04.008
Author Profile
ACKNOWLEDGEMENTS
The authors are immensely grateful to the valuable
suggestions and help provided by the Department of
Electronics and Communication Engineering, KLE
Dr. M.S. Sheshagiri College of Engineering and
Technology, Udyambag, Belgaum, Karnataka.
Prof.Sheela Shankar has completed
her Bachelor of Engineering in Electronics and
Communication from BIET, Davangere, Karnataka.
She has pursued her Masters in Electronics and
Control Engineering from Birla Institute of
Technology and Science, Pilani. Currently she is
working as an associate professor in the department
of Electronics and Communication Engineering, KLE
Dr.M.S.Sheshgiri College of Engineering and
Technology, Belgaum, Karnataka, India. Her areas of
research includes image processing, communication
engineering and control engineering.
REFERENCES
[1]
Xiaozheng Zhang, YongshengGao, “Face recognition across
pose: A review”, Pattern Recognition 42 (2009) 2876 –
2896, Elsevier. doi: 10.1016/j.patcog.2009.04.017
[2]
RabiaJafri, Hamid R. Arabnia, “A Survey of Face
Recognition Techniques”, Journal of Information
Processing Systems, Vol.5, No.2, June 2009. DOI:
10.3745/JIPS.2009.5.2.041
Proceedings of 10th IRF International Conference, 04th October-2014, Bengaluru, India, ISBN: 978-93-84209-56-8
58
Using Eigenface Features to Drive Support Vector Machines in Face Recognition Systems
Maharashtra state, in 2003. His field of interests
includes signal processing, Image processing,
cryptography, and knowledge based systems.
Currently he is working as a professor in Electronic
and communication department of Gogte institute of
Technology, Belgaum, Karnataka state.
Dr. V. R. Udupidid his bachelor’s
degree in Electronics and communication Engg. from
Mysore University in 1984 and pursued his master’s
degree in Electronics Engineering with computer
applications as specialization from Shivaji University,
Kolhapur, Maharashtra state, in 1989.
He has 30 years of total teaching experience and
currently he is guiding 05 research scholars and has
guided 04 candidates for Ph.D. He has published
more than 42 technical papers in national and
international conferences and 08 articles in journals.
He is a life member of ISOI, SSI, CSI, BMESI, and
ISTE.
He has completed his doctoral degree in Electrical
Engineering from Shivaji University, Kolhapur,

Proceedings of 10th IRF International Conference, 04th October-2014, Bengaluru, India, ISBN: 978-93-84209-56-8
59

Download Report