JOURNAL OF COMPUTING, VOLUME 2, ISSUE 7, JULY 2010, ISSN 2151-9617

HTTPS://SITES.GOOGLE.COM/SITE/JOURNALOFCOMPUTING/
WWW.JOURNALOFCOMPUTING.ORG
6

Rotation Independent Face Recognition using

Neuro-Genetic Hybrid Algorithm
Md. Rabiul Islam, Md. Fayzur Rahman

Abstract—This paper presents an approach to enhance the performance of a face recognition system using hybrid algorithm
based on Artificial Neural Networks (ANN) optimized by Genetic Algorithm (GA) introducing sustainability of rotational distortion.
Though the traditional face recognition system is very sensitive to the face parameter variations, the proposed Neuro-Genetic
hybrid system is found to be stance and performs well for improving the robustness and naturalness of human-computer-
interaction. In this work, we investigated two approaches in order to improve face recognition performance for the rotational face
environment: one seeks to extract the features from the face image by using improved and efficient image pre-processing
techniques. Fuzzy membership function has been used for the feature extraction purpose. The other task combines the
extracted features that have been used by the Neuro-Genetic hybrid algorithm to improve the performance of rotational face
recognition system. Experimental results show the superiority of the proposed rotational independent face recognition system
according to various orientations.

Index Terms— Human Computer Interaction, Rotational Face Recognition, Neuro-Genetic Hybrid Algorithm, Fuzzy
Membership Function, Facial Feature Extraction.

——————————  ——————————

1 INTRODUCTION

B iometrics are seen by many researchers as a solution to a

lot of user identification and security problems now a
days [1]. Automatic recognition of person has always
been an attractive goal of informatics. Face recognition based
personal identification results in a highly accurate and se-
cured system that has been widely applied in various prac-
tical fields [2], [3], [4].
Several excellent survey papers on face recognition tech-
niques are available with a wide variety of methods [5], [6],
[7], [8] that covers early face recognition approaches. While
humans quickly and easily recognize faces under variable
situations or even after several years of separation, the prob-
lem of machine face recognition is still a highly challenging
task in pattern recognition and computer vision [9], [10]. A
face is inherently a 3D object illuminated by a variety of light-
ing sources from different directions and surrounded by arbi-
trary background objects. Therefore, the appearance of a face
varies tremendously when projected onto a 2D image. Differ-
ent pose angles cause significant changes in 2D appearance
[11]. Rotation independent face recognition using optical
neural network has also been developed [12], [13].
Various learning algorithm such as probabilistic deci-
sion-based neural network based face recognition/detection
[14], neural network based face detection [15], Gabor Wave-
lets Transform and Extended Nearest Feature Space Classifi-
cation [16], Karhunen-Loeve Procedure [17], ICA based face
detection [18], support vector machine based approach [19],
————————————————
 Md. Rabiul Islam is with the Department of Computer Science & Engi-
neering, Rajshahi University of Engineering & Technology, Rajshahi-6204,
Bangladesh.
 Md. Fayzur Rahman is with the Department of Electrical & Electroic En-
gineering, Rajshahi University of Engineering & Technology, Rajshahi-
6204, Bangladesh.
eigenfaces for recognition [20] etc. have been developed for
the face recognition purpose. Learning systems in face rec-
ognition that employ hybrid strategies [21], [22], [23], [24]
can potentially offer significant advantages over single-
strategy systems.
In this proposed system, neural network (NN) and genet-
ic algorithm (GA) has been used as a hybrid algorithm. To
extract the features from the face, a new approach of feature
extraction technique has been adopted for the rotation inde-
pendent face recognition system. The individual perfor-
mance of the NN and GA has been simulated and finally the
overall performance of this proposed hybrid system has
been populated according to the VALID face database.
2 PROPOSED TECHNIQUE FOR THE FACE
RECOGNITION SYSTEM
The proposed technique is based on the idea that each point
on a circle (excluding inside or outside of a circle) has a static
distance from its centroid. This distance is called the radius
(r). A point can be moved anywhere on a circle but the dis-
tance ‘r’ from its centroid will not be changed. Also the
number of points is static and will never be changed when
they are moved on a circle. Of course the number of points
will be changed if we do not have a complete circle. This is
the case of distorted images. One of the main roles of fuzzy
techniques which we have used in our approach is to make a
complete circle from an incomplete one. From the shape and
appearance based features, appearance based feature extrac-
tion technique has been used in this work. Neuro- Genetic
hybrid algorithm has been used for learning and classifica-
tion of data collected from previous feature extraction stage.
The paradigm of this work is shown in Fig. 1.
 JOURNAL OF COMPUTING, VOLUME 2, ISSUE 7, JULY 2010, ISSN 2151-9617
HTTPS://SITES.GOOGLE.COM/SITE/JOURNALOFCOMPUTING/
WWW.JOURNALOFCOMPUTING.ORG
Fig.1 Paradigm of the proposed rotation independent face recogni-
tion system using hybrid technique.
3 PROPOSED TECHNIQUE FOR THE FACE
RECOGNITION SYSTEM
The first step in image pre-processing is image acquisi-
tion. To do so, an imaging sensor along with signal digiti-
zation capability has been used so that captured image
can be converted to digital form directly. After acquisition
of face image, Stams [25] Active Appearance Model
(ASM) has been used to detec the facial features. Then the
binary image has been taken. The Region Of Interest
(ROI) has been choosen according to the selection ROI
[26], [27]. Lastly the background noise has been eliminat-
ed [28] and finally appearance based facial feature has
been found. The procedure of the facial image pre-
processig parts are shown in Fig. 2.
Fig. 2. Facial image pre-processig for the proposed system (a)
original image (b) Output taken from Stams Active Appearance
Model (c) Facial edges are extracted (d) Region Of Interest (ROI)
selection with background noise (e) Appearance based facial
features.
4 FEATURE EXTRACTION FROM THE FACIAL IMAGE
Feature extraction extracts the recognizable properties of
an image. To do this segmentation is required. Segmenta-
tion partitions an input image into its constituent parts or
objects [28]. To extract the feature of face images [29], at
first some blocks are created on the whole image, then
values from those blocks (one value from each block) are
calculated. A threshold has been used and finally a binary
matrix has been extracted [30].
7
Reduced form of binary matrix is essential for reduc-
ing the time complexity of the Neuro-Genetic hybrid sys-
tem. From the reduced binary matrix two types of infor-
mation has been found. One is pixels 0 (white) and anoth-
er is 1 (black). After completion of the conversion, unne-
cessary pixels from the original image are removed using
the following algorithm:
Algorithm for calculating the binary reduced matrix
from the image:
BEGIN
Step1. Start from top-left corner, repeat for each column
and row
If (sum of all black pixels) in column/row > 0
then save column and row
else don’t save (delete from the face image) col
umn/row
Step2. Calculate and save height and width of the re-
duced binary form of face image
END
We calculate the Center of the preprocessed image us-
ing the formula as below.
Width  1 Height  1                 (1) 
Center_x  ; Center_y 
2 2
Once we have got the center of the preprocessed face
image, we have drawn circles with centroid (Center_x,
Center_y) and various radii (r_max, ... ,r_min) as shown
in Fig.3 and the sum of all black pixels as well as all black
and white pixels (total pixels) belonging to each circle
have calculated. This sum is fuzzy set for distorted face
images. We have used fuzzy sets with a membership
function as shown in Fig. 4 and we have calculated the
membership values for each radius and those values are
used by the neuro-genetic hybrid system for learning and
classification. For this purpose we have used the follow-
ing algorithm:
Algorithm for using the fuzzy membership function
for feature extraction:
BEGIN
Step1. For each black pixel (x,y), calculate radius (r) using
the following formula:
r center _ x  x   center _ y  y 
2
         (2) 
2
Step2. Sum all pixels with same radius
(sum_r_max,……......,sum_r_min)
Step3. Calculate the membership for each circle or disk
using the following membership function:
 bp rbp 
m _ f bp   e  total _ pixels                                                   (3) 
END
Here, bp = sum of black pixels (testing face images),
rbp = sum of real black pixels (average taken from many
training face images), total_pixels = (black + white) pixels
belong to circle. If (bp = rbp) then exp(0) is 1 and function
m_f(bp) gets peak value (1). It means that the circle from
current image is the same as circles from face images used
JOURNAL OF COMPUTING, VOLUME 2, ISSUE 7, JULY 2010, ISSN 2151-9617
HTTPS://SITES.GOOGLE.COM/SITE/JOURNALOFCOMPUTING/
WWW.JOURNALOFCOMPUTING.ORG 8

for training. Generate the initial population Pi of real coded chromo-

If bp < rbp and/or bp > rbp, the function value somes C i j each representing a weight set for the BPN;
m_f(bp) goes away from 1. If the value is 0.9, it means Generate fitness values F i j for each C i j € Pi using the
that the circle from the current face image is nearly the algorithm FITGEN();
same as the circle from trained face images. If the value is While the current population Pi has not converged
less than 0.5, it means that the circle is not very similar to {
those used for training. Using the cross over mechanism reproduced offspring
from the parent chromosome and performs mutation on
offspring;
i i+1;
Call the current population Pi ;
Calculate fitness values F i j for each C i j € Pi using the
algorithm FITGEN();
}
Fig. 3 Processing techniques of face image creating circles using Extract weight from Pi to be used by the BPN;
centroid and radius. }
Algortithm for FITGEN():
{
I , Tj Ii I ,
Let ( i
I 2i , I li), i=1,2,………..N
Ti T1i where
, T2i , = ( 1i
Tli
……… ) and = ( ……… )
represent the input-output pairs of the problem to be
solved by BPN with a configuration l-m-n.
{
W
Fig.4 Membership function for creating input to Neuro-Genetic Extract weights i from Ci ;
hybrid algorithm. Wi
Keeping as a fixed weight setting, train the BPN for
the N input instances (Pattern);
5 TRAINING AND TESTING MODEL OF THE NEURO- Calculate error Ei for each of the input instances using the
GNETIC HYBRID TECHNIQUE formula:
 
Fig.5 shows the working process of neuro-genetic hybrid
system [31], [32], [33]. The structure of the multilayer Ei, =   (T ji  O ji ) 2                                                            (4) 
neural network does not matter for the GA as long as the j
BPNs parameters are mapped correctly to the genes of the Where Oi  is the output vector calculated by BPN; 
chromosome the GA is optimizing. Basically, each gene
represents the value of a certain weight in the BPN and Find the root mean square E of the errors Ei,
the chromosome is a vector that contains these values Where i = 1,2,…….N
such that each weight corresponds to a fixed position in
the vector as shown in Fig.5.
i.e. E =  E i
i N ;                                                           (5) 
The fitness function can be assigned from the recogni- Now the fitness value Fi for each of the individual string
tion error of the BPN for the set of pictures used for train- of the population as Fi = E;
ing. The GA searches for parameter values that minimize }
the fitness function, thus the recognition error of the BPN Output Fi for each Ci, i = 1,2,…….P; }
is reduced and the recognition rate is maximized [34]. }

6 OPTIMUM PARAMETER SELECTION FOR THE

Fig.5 Learning and recognition model for the Neuro-Genetic hybr-
id system.
The algorithm for the GA-NN based weight determi-
nation and Fitness Function [35] is as follows:
Algorithm for GA-NN Weight determination:
{  
i 0;
BPN AND GA
There are some critical parameters in Neuro-Genetic hy-
brid system (such as in BPN, gain term, speed factor,
number of hidden layer nodes and in GA, crossover rate
and the number of generation) that affect the performance
of the proposed system. A trade off is made to explore the
optimal values of the above parameters and experiments
are performed using those parameters. The optimal val-
ues of the above parameters are chosen carefully and fi-
nally the following recognition rate is found.
6.1 Experiment on the Gain Term,  for BPN
In BPN, during the training session when the gain term
was set as: 1 = 2 = 0.3, spread factor was k1 = k2 = 0.20
JOURNAL OF COMPUTING, VOLUME 2, ISSUE 7, JULY 2010, ISSN 2151-9617
HTTPS://SITES.GOOGLE.COM/SITE/JOURNALOFCOMPUTING/
WWW.JOURNALOFCOMPUTING.ORG 9

and tolerable error rate was fixed to 0.001% then the

highest identification rate of 92 % has been achieved at
gain term 0.3 which is shown in Fig.6.

Fig.9 Performance measurement according to the crossover rate.

6.5 Experiment on the Number of Generations for

GA
Fig.6 Performance measurement according to gain term.
Different values of the number of generations have been
tested for achieving the optimum number of generations.
6.2 Experiment on the Speed Factor, k for BPN The test results are shown in the Fig.10. The maximum
The performance of the BPN system has been measured identification rate of 96 % has been found at the number
 
according to the speed factor, k. We have set 1 = 2 = of generations 5.
0.3 and tolerable error rate was fixed to 0.001%. We have
studied the value of the parameter ranging from 0.1 to
0.5. We have found that the highest recognition rate was
94 % at k1 = k2 = 0.15 which is shown in Fig.7.

Fig.10 Performance according to the no. of generations.

7 EXPERIMENTAL RESULTS AND PERFORMANCE

Fig.7 Performance measurement according to speed factors.
6.3 Experiment on the Number of Nodes in Hidden
State for BPN
In the learning phase of BPN, We have chosen the hidden
layer nodes in the range from 5 to 40. We have set 1 =
 evaluate the performance, the test face orientation was
 2 = 0.3, k1 = k2 = 0.15 and tolerable error rate was fixed
to 0.001%. The highest recognition rate of 95 % has been
achieved at NH =35 which is shown in Fig.8.
ANALYSIS
VALID face database [36] has been used to measure the per-
formance of the proposed hybrid system. Fig.11 shows some
processed face images from VALID database. In learning
phase, face images with 0 rotation were used. In order to
changed from 0 to 90 ans 270 to 360 with the increments
of 10. Tabel 1 shows the performance results of the pro-
posed rotation independent Neuro-Genetic hybrid system
according to the VALID database. From the results it is
shown that the recognition rate decreases with the increment
of the rotation distortion but in the identical case i.e. on 0
and 360 the recognition rate is 100%.

Fig. 8 Results after set up the number of internal nodes in BPN.

6.4 Experiment on the Crossover Rate for GA

In this experiment, crossover rate has been changed in
various ways such as 1, 2, 5, 7, 8, 10. The highest speaker
identification rate of 95 % was found at crossover point 5
Fig.11 Some processed facial images for the proposed system.
which is shown in the Fig.9.
JOURNAL OF COMPUTING, VOLUME 2, ISSUE 7, JULY 2010, ISSN 2151-9617
HTTPS://SITES.GOOGLE.COM/SITE/JOURNALOFCOMPUTING/
WWW.JOURNALOFCOMPUTING.ORG
TABLE 1
PERFORMANCE MEASUREMENT OF THE PROPOSED SYSTEM
8 CONCLUSION AND OBSERVATION
The experimental results show the versatility of the Neu-
ro-Genetic hybrid algorithm based rotation independent
face recognition system. The critical parameters such as
gain term, speed factor, number of hidden layer nodes,
crossover rate and the number of generations have a great
impact on the recognition performance of the proposed
system. The optimum values of the above parameters
have been selected effectively to find out the best perfor-
mance. The highest recognition rate of BPN and GA has
been achieved at 95 % and 96 % respectively. According
to the VALID database, Table 1 shows the recognition
accuracy that has been achieved in Neuro-Genetic hybrid
system with different orientations. Therefore, this pro-
posed system can be used in various security and access
control purposes. The performance of the system can be
improved by using efficient image pre-processing tech-
niques. Finally the performance of this proposed system
can be populated according to the largest face recognition
database.
REFERENCES
[1] Jain, A., Bole, R., and Pankanti, S., BIOMETRICS Personal Identi-
fication in Networked Society. Kluwer Academic Press, Boston,
1999.
[2] S. Akamatsu, “The research trend of face recognition by com-
puter,” The Journal of IEICE, Vol.80, No.3, pp. 257-266, 1997.
[3] O. Hasegawa, S. Morisima, and M. Kameko, “Information
processing of face,” IEICE Trans. D-II, Vol.80, No.8, pp. 2047-
2065, 1997.
[4] M. Yachida, Robot vision. Syokoudou Publishing, 1993.
[5] A. Samal and P.A. Iyengar, “Automatic recognition and analy-
sis of human faces and facial expressions: a survey,” Patt. Re-
cogn. 25 (1) (1992), pp. 65–77, 1992.
[6] D. Valentin, H. Abdi, A.J. O_Toole, “G.W. Cottrell, Connection-
ist models of face processing: a survey,” Patt. Recogn. 27 (9)
(1994) pp. 1209–1230, 1994.
[7] R. Chellappa, C.L. Wilson and S. Sirohey, “Human and ma-
chine recognition of faces: a survey,” Proc. IEEE 83 (5) (1995)
pp. 705–740, 1995.
[8] J. Zhang, Y. Yan and M. Lades, “Face recognition: eigenface,
elastic matching, and neural nets,” Proc. IEEE 85 (9) (1997) pp.
1423–1435, 1997.
[9] I. Craw, N. Costen, T. Kato and S. Akamatsu, “How should we
represent faces for automatic recognition?,” IEEE Trans. Patt.
Anal. Mach. Intell. 21 (8) (1999), pp. 725–736, 1999.
10
[10] A.M. Burton, V. Bruce, P.J.B. Hancock, “From pixels to people:
a model of familiar face recognition,” Cognitive Sci. 23 (1999),
pp. 1–31, 1999.
[11] S. Kong, J. Heo, B. Abidi, J. Paik and M. Abidi, “Recent Ad-
vances in Visual and Infrared Face Recognition - A Review,”
The Journal of Computer Vision and Image Understanding, Vol. 97,
No. 1, pp. 103-135, 2005.
[12] K. Parimala Geetha, S. Sundaravadivelu and N. Albert Singh,
“Rotation Invariant Face Recognition using Optical Neural
Networks,” TENCON 2008 - 2008 IEEE Region 10 Conference,
Hyderabad, India, 2009.
[13] Kiyomi Nakamura, and Hironobu Takano, “Rotation and Size
Independent Face Recognition by the Spreading Associative
Neural Network,”International Joint Conference on Neural Net-
works, Sheraton Vancouver Wall Centre Hotel, Vancouver, BC,
Canada, 2006.
[14] S. H. Lin, S. Y. Kung, and L. J. Lin., “Face recognition/detection
by probabilistic decision-based neural network,” IEEE Transac-
tions on Neural Networks, Special Issue on Artificial Neural Net-
works and Pattern Recognition, 8(1), 1997.
[15] Henry A. Rowley, Shumeet Baluja, and Takeo Kanade, 1998.
Neural network based face detection. IEEE Transactions on Pat-
tern Analysis and Machine Intelligence, 20(1).
[16] Jianke Zhu, Mang I Vai and Peng Un Mak, ”Gabor Wavelets
Transform and Extended Nearest Feature Space Classifier for
Face Recognition,” Proceedings of the Third IEEE International
Conference on Image and Graphics (ICIG’04), 2004.
[17] M. Kirby and L. Sirovich, “Application of the Karhunen-Loeve
Procedure for the Characterization of Human Faces,” IEEE 
Trans. Pattern Analysis and Machine Intelligence, vol. 12, no. 1, pp. 
103‐108, 1990. 
[18] J. Karhunen, E. Oja, L. Wang, R. Vigario, and J. Joutsensalo, “ A 
class of neural networks for independent component analysis,” 
IEEE Trans. Neural Networks, vol. 8, pp. 486–504, 1997. 
[19] E.  Osuna,  R.  Freund,  and  F.  Girosi,  “Training  support  vector 
machines:  An  application  to  face  detection,”  Proc.  IEEE  Conf. 
Computer Vision and Pattern Recognition, pp. 130–136 1997. 
[20] M. Turk and A. Pentland, “Eigenfaces for recognition,” J. Cogni‐
tive Neuroscience, vol. 3, no. 1, pp. 71–86, 1991. 
[21] Y.‐S. Ryu and S.‐Y. Oh, “Simple hybrid classifier for face recog‐
nition with adaptively generated virtual data,” Pattern Recognit. 
Leters., vol. 23, pp. 833–841, 2002. 
[22] J.  Bala,  P.  Pachowicz,  and  K.  De  Jong,  “Multistrategy  Learning 
from Engineering Data by Integrating Inductive Generalization 
and Genetic Algorithms, in Machine Learning: A Multistrategy 
Approach,”  Michalski  and  G.  Tecuci  (Eds.),  Vol.  IV,  R.S.  Morgan 
Kaufmann, San Mateo, CA., pp. 121‐138, 1994. 
[23] F.  Gruau  and  D.  Whitley,  “Adding  Learning  to  the  Cellular 
Development  of  Neural  Networks:  Evolution  and  the  Baldwin 
Effect,” Evolutionary Computation, Vol.1, No.3, pp. 213‐234, 1993. 
[24] H. Vafaie and K. De Jong, “Improving a Rule Induction System 
Using Genetic Algorithms, in Machine Learning: A Multistrate‐
gy Approach,” Michalski and G. Tecuci (Eds.), Vol. IV, R.S., Mor‐
gan Kaufmann, San Mateo, CA., pp. 453‐469, 1994. 
[25] Stephen Milborrow and Fred Nicolls, “Locating Facial Features 
with  an  Extended  Active  Shape  Model,”  available  at 
http://www.milbo.org/stasm‐files/locating‐facial‐features‐with‐
an‐extended‐asm.pdf. 
[26] R. Herpers, G. Verghese, K. Derpains and R. McCready, “Detec‐
tion and tracking of face in real environments,” IEEE Int. Work‐
JOURNAL OF COMPUTING, VOLUME 2, ISSUE 7, JULY 2010, ISSN 2151-9617
HTTPS://SITES.GOOGLE.COM/SITE/JOURNALOFCOMPUTING/
WWW.JOURNALOFCOMPUTING.ORG 11

shop  on  Recognition,  Analysis  and  Tracking  of  Face  and  Gesture  in  Md. Fayzur Rahman was born in 1960
in Thakurgaon, Bangladesh. He re-
Real‐ Time Systems, Corfu, Greece, pp. 96‐104, 1999. 
ceived the B. Sc. Engineering degree in
[27] J. Daugman, “Face detection: a survey,” Comput.  Vis. Imag.  Un‐ Electrical & Electronic Engineering from
derst, 83, 3, pp.  236‐ 274, 2001.  Rajshahi Engineering College, Bangla-
[28] Rafael  C.  Gonzalez  and  Richard  E.  Woods,  Digital  Image  desh in 1984 and M. Tech degree in
Industrial Electronics from S. J. College
Processing. Addison‐Wesley, 2002. 
of Engineering, Mysore, India in 1992.
[29] K. Wong, K. Lam and W. Siu, “An efficient algorithm for human  He received the Ph. D. degree in ener-
face detection and facial feature extraction under different con‐ gy and environment electromagnetic
ditions,” Pattern Recognition, Vol. 34, 2001.  from Yeungnam University, South Korea, in 2000. Following his
graduation he joined again in his previous job in BIT Rajshahi. Cur-
[30] Kaushik  Roy,  Abu  Sasyeed  Md.  Sohail,  Md.  Rabiul  Islam  and 
rently he is a Professor in Electrical & Electronic Engineering in Raj-
A.H.M.  Sarower  Satter,  “Image  Processing  Techniques  for  real  shahi University of Engineering & Technology (RUET). He is current-
Time Eye Recognition,” 2nd International Conference on Computer  ly engaged in education in the area of Electronics & Machine Control
Science and Its Application, National University, San Diego, Cali‐ and Digital signal processing. He is a member of the Institution of
Engineer’s (IEB), Bangladesh, Korean Institute of Illuminating and
fornia, USA, 2004 
Installation Engineers (KIIEE), and Korean Institute of Electrical En-
[31] Siddique and M. & Tokhi, M., “Training Neural Networks: Back  gineers (KIEE), Korea.
Propagation vs. Genetic Algorithms,” Proceedings of International 
Joint Conference on Neural Networks, pp. 2673‐ 2678, Washington 
D.C.USA,2001  
[32] Whiteley,  D.,  “Applying  Genetic  Algorithms  to  Neural  Net‐
works Learning,” Proceedings of Conference of the Society of Artifi‐
cial Intelligence and Simulation of Behavior, pp. 137‐ 144, England, 
Pitman Publishing, Sussex, 1989. 
[33] Whiteley, D., Starkweather and T. & Bogart, C., “Genetic Algo‐
rithms  and  Neural  Networks:  Optimizing  Connection  and 
Connectivity,” Parallel Computing, Vol. 14, pp. 347‐361, 1990. 
[34] Kresimir  Delac,  Mislav  Grgic  and  Marian  Stewart  Bartlett,  
Recent  Advances  in  Face  Recognition.  I‐Tech  Education  and  Pub‐
lishing KG, pp. 223‐246, Vienna, Austria, 2008. 
[35] Rajesskaran  S.  and  Vijayalakshmi  Pai,  G.A.,  Neural  Networks, 
Fuzzy  Logic,  and  Genetic  Algorithms‐  Synthesis  and  Applications. 
Prentice‐Hall of India Private Limited, New Delhi, India, 2003. 
[36] N. A. Fox, B. A. OʹMullane and R. B. Reilly, “The Realistic Mul‐
ti‐modal  VALID  database  and  Visual  Speaker  Identification 
Comparison  Experiments,”  Proc.  of  the  5th  International  Confe‐
rence  on  Audio‐  and  Video‐Based  Biometric  Person  Authentication 
(AVBPA‐2005), New York, 2005. 

Md. Rabiul Islam was born in Rajshahi,

Bangladesh, on December 26, 1981. He
received his B.Sc. degree in Computer
Science & Engineering and M.Sc. de-
grees in Electrical & Electronic Engineer-
ing in 2004, 2008, respectively from the
Rajshahi University of Engineering &
Technology, Bangladesh. From 2005 to
2008, he was a Lecturer in the Depart-
ment of Computer Science & Engineering at Rajshahi University of
Engineering & Technology. Since 2008, he has been an Assistant
Professor in the Computer Science & Engineering Department, Uni-
versity of Rajshahi University of Engineering & Technology, Bangla-
desh. He is now pursuing his Ph.D. degree programme under the
excellent guidance of Dr. Md. Fayzur Rahman, Department of EEE
at Rajshahi University of Engineering & Technology, Bangladesh.
His research interests include bio-informatics, human-computer inte-
raction, speaker identification and authentication under the neutral
and noisy environments.