Académique Documents
Professionnel Documents
Culture Documents
A
i
C
j
k k
j i B
AXC
E
1 1
) , (
1
(1)
AXC is the size of the image block.
Neutral Network training algorithm
Ear image decomposition using Haar wavelet
transforms
Features extraction from LH
1
,
HL
1
and HH
1
image-blocks
Fusion of image energy and edge density features
Neural Network testing algorithm
LL
1
1LH
1
2LH
1
3LH
1
4LH
1
1HL
1
2HL
1
3HL
1
4HL
1
1HH
1
2HH
1
3HH
1
4HH
1
International Journal of Video & Image Processing and Network Security IJVIPNS-IJENS Vol: 11 No: 01 28
118001-3232 IJVIPNS-IJENS February 2011 IJENS
I J E NS
(3) Calculate the edge density feature of each of image
blocks. This feature measures the average edge magnitudes in
the region. The edge magnitude is a combination of the edge
strengths along the horizontal and vertical directions . The
edge magnitude image (M(i, j)) is found by applying edge
detector called Sobel operator to B(i j). Two sober edge
detector operators (G
x
and Gy) are used to calculate the
horizontal and vertical gradient of image intensity at each
point respectively using (2).
x
B
G
x
and
y
B
G
y
(2)
The gradient magnitude M(i,j) is calculated using (3)
2 2
) , (
y x
G G j i M
(3)
For each of image block B (i j), the edge density feature (D
k
)
is given as:
A
i
C
j
k
j i M
AXC
D
1 1
) , (
1
(4)
(4) Fuse the two features together to form a short-length
feature vector. The resulting combined energy-edge density
feature vector is given as:
n n ed
D D D D D E E E E E f . .......... , , , , ......... , , ,
4 3 2 1 4 3 2 1
(5)
Where n is the number of image blocks.
(5) Normalize the fused feature vector (f
ed
) to values that
range from 0 to 1. This is achieved by dividing each of the
feature value by the highest feature component.
IV. BACKPROPAGATION NEURAL NETWORK FOR
EAR RECOGNITION
Back Propagation Neural Network (BPNN) is a multilayered,
feed forward Neural Network (NN) and is by far the most
widely used. The basic structure of the BPNN includes one
input layer, at least one hidden layer (there could be a single
layer or multiple layers), followed by output layer. The
network receives inputs information by neurons in the input
layer and the output information is given by the neurons in the
output layer. This allows the network to learn non linear and
linear relationships between input and output vectors. This
means that the interconnected neurons are organized in layers
and send their signals forward and then the errors are
propagated backward. Back Propagation works by adjusting
the weight values during training in order to reduce the error
between actual and desire output pattern [11].
In this work, we used three layers BPNN as the classifier,
the number of node in the input layer is equal to dimension of
the feature vector that characterizes the ear image
information. The number of the node in the hidden layer is set
by trial and error during training. The number of node in the
out layer is equal to the number of the subjects in the
database.
During learning process, the number of node in the hidden
layer, the learning coefficient and the momentum rate are
adjusted in order to achieved the set minimum error value
which is consider suitable for this application, once the Mean
Square Error (MSE) is reached, the final weight and the
threshold of the network are stored to create knowledge base
for the recognition stage. The BPNN learning algorithm tries
to minimize the difference MSE between the targets and
predicts output by adjusting weight and bias of each of the
neurons. The total MSE is computed as given in (6) using all
training images.
) (
2
1
1 1
ij
m
j
k
i
ij
O T MSE
(6)
Where m is the number of training sample, k is the number of
output units, T
ij
is the target output value of the ith output unit
for the jth training sample, and O
ij
is the actual output of the
ith output unit for the jth training sample.
In the recognition phase, the features from the query
ear image that is to be tested is fed into the network without
having any target output, BPNN testing algorithm found the
closest matching output using the weights and thresholds that
have stored and provided the corresponding recognized ear.
V. EXPERIMENTAL RESULTS
Experiments have been conducted to evaluate the
effectiveness of the system by using our ear images database.
The database contains 1050 ear images collected from 350
subjects. We have trained the BRNN on all subjects present in
the database. We have provided the network with 2 training
images of each of the subjects. We have calculated the
network response recognition rate with different number of
subjects as shown in Table I. Also in Table II, the recognition
rates of the system based on Euclidean distance and Neural
Network is compared. The results from TableII showed that
the recognition rates of the method proposed using BPNN is
higher in comparison with other previous method proposed in
[7 ][ 8].
International Journal of Video & Image Processing and Network Security IJVIPNS-IJENS Vol: 11 No: 01 29
118001-3232 IJVIPNS-IJENS February 2011 IJENS
I J E NS
TABLE I
NEURAL NETWORK RECOGNITION RESPONSE
Number
of
Subject
Recognized
ear image
Unrecognized
ear image
Recognition
rate (%)
50 49 1 98.00
150 147 3 98.00
250 243 7 97.20
350 341 9 97.42
TABLE II
RECOGNITION RATE OF BACK-PROPAGATION NEURAL NETWORK AND
EUCLIDEAN DISTANCE
Classifier
Num-
ber of
Subject
Recognized
ear image
Unrecognized
ear image
Recognit
ion rate
(%)
Euclidean
Distance
150 130 20 86.67
BPNN 150 147 3 98.00
VI. CONCLUSION
In this paper, we have introduced a new approach for ear
recognition for better security application. The high
recognition rate results obtained can be traced to combination
strength of Back Propagation Neural Network and fused
feature used in the proposed system. Wavelet transforms and
blocks division methods have helped to access high frequency
textural features of the ear image at local level.
REFERENCES
[1] B. Vict or, K. Bowyer, and s. Sarkar, An Evaluat ion of face and
Ear Biomet ric, Proc. 16
th
Int ernat ional conference of Pat t ern
Recognit ion, 2002, pp.429-432.
[2] K. chang, K.Bowyer, and V. Barnabas, Comparison and
Combinat ion of Ear and Face Image in Appearance-Based
Biomet rics, IEEE Trans. Pat t ern Analysis and Machine
Int elligence, 2003, 25, pp. 1160-1165.
[3] A. A. Darwish, R. Abd Elghafar and A. Fawzi Ali, Mult imodal
Face and Ear Images, Journal of Comput er Science 5(5), 2009
pp374-379.
[4] D. Hurley, M. Nixon and Cart er, Force Field Energy Funct ional
for Ear Biometrics, Computer Vision and Image Underst anding,
2005, 98, pp.491-512.
[5] M. Burge and W. Burger, Ear biometrics in Computer Vision, in
Proceedings of t he 15
th
Int ernat ional Conference on Pat t ern
recognit ion, 2000.
[6] Md. Mahbubur Rahman, Md. Rashedul Islam, Nazmul Islam
Bhuiyan, Bulbul Ahmed, Md. Aminul Islam , Person
Ident ification Using Ear Biometrics International Journal of The
Comput er, t he Int ernet and Management Vol. 15: 2 (May -
August , 2007) pp 1 8.
[7] Y. Wang, Z. Mu, and H. Zeng, Block-based and mult i-resolut ion
met hods for ear recognition using wavelet t ransform and uniform
local binary pat t erns, Proceedings of t he 19
th
Int ernat ional
Conference on Pat t ern recognit ion, 2008.
[8] Anupam Sana, Phalguni Gupt a and Ruma Purkait, Ear biometric:
A New Approach, WSPC- Proceedings (ICAPR 2007).
[9] S. Arivazhagan L. Ganesan Aut omat ic Target Det ect ion Using
Wavelet Transform EURASIP Journal on Applied Signal
Processing 2004:17, 26632674.
[10] U.S.N Raju, A. Srikrishna, V. Vijaya Kumar and A. Suresh,
Ext ract ion of Skelet on Primit ives on Wavelet s, Journal of
Theoretical and Applied Information Technology pp1065 1074,
2008.
[11] S.A Daramola, Development of a Robust Aut omatic Off-line and
On-line Handwrit ten Signat ure Verification Syst em," PhD. Thesis,
Depart ment of Electrical and Information Engineering Covenant
Universit y Ot a. June, 2008.
Authors Profile
Dr. Samuel Adebayo Daramol a obt ained Bachelor of Engineering from
Universit y of Ado-Ekit i Nigeria, in 1997. Mast er of Engineering from
Universit y of Port harcourt , Nigeria in 2002 and PhD from Covenant
Universit y, Nigeria in 2008. He is working current ly as a Lect urer for t he
Elect rical and Information Engineering department in Covenant Universit y
Ot a Nigeria. He has published many papers in int ernat ional Journals. His
research int erest s include Image processing and crypt ography.
Ol adejo Daniel Oluwaninyo is a graduat ing st udent in t he depart ment of
Elect rical and Information Engineering Covenant Universit y Ot a, Nigeria
2011. His research interests include Image processing and communicat ion.