An Efficient Fingerprint Identification using Neural Network and BAT Algorithm

International Journal of Electrical and Computer Engineering (IJECE)
Vol. 8, No. 2, April 2018, pp. 1194~1213
ISSN: 2088-8708, DOI: 10.11591/ijece.v8i2.pp1194-1213  1194
Journal homepage: http://iaescore.com/journals/index.php/IJECE
An Efficient Fingerprint Identification using Neural Network
and BAT Algorithm
Subba Reddy Borra1
, G. Jagadeeswar Reddy2
, E.Sreenivasa Reddy3
1
Department of Computer Science and Engineering, JNTUH, Hyderabad, 500085, India
2
Vaagdevi Institute of Technology and Sciences, Proddatur, AP, 516360, India
3
University College of Engineering, Acharya Nagarjuna University, AP, 522508, India
Article Info ABSTRACT
Article history:
Received Nov 15, 2017
Revised Dec 27, 2017
Accepted Jan 5, 2018
The uniqueness, firmness, public recognition, and its minimum risk of
intrusion made fingerprint is an expansively used personal authentication
metrics. Fingerprint technology is a biometric technique used to distinguish
persons based on their physical traits. Fingerprint based authentication
schemes are becoming increasingly common and usage of these in
fingerprint security schemes, made an objective to the attackers. The repute
of the fingerprint image controls the sturdiness of a fingerprint authentication
system. We intend for an effective method for fingerprint classification with
the help of soft computing methods. The proposed classification scheme is
classified into three phases. The first phase is preprocessing in which the
fingerprint images are enhanced by employing median filters. After noise
removal histogram equalization is achieved for augmenting the images. The
second stage is the feature Extraction phase in which numerous image
features such as Area, SURF, holo entropy, and SIFT features are extracted.
The final phase is classification using hybrid Neural for classification of
fingerprint as fake or original. The neural network is unified with BAT
algorithm for optimizing the weight factor.
Keyword:
BAT algorithm
Classification
Histogram equalization
Hybridneural network
Median filter
Copyright © 2018 Institute of Advanced Engineering and Science.
All rights reserved.
Corresponding Author:
Subba Reddy Borra ,
Department of Computer Science and Engineering,
JNTUH,
Hyderabad, 500085, India.
Email: bvsr79@gmail.com
1. INTRODUCTION
Multi-biometric recognition [1] and biometric template protection [2] are defined in this work to
derive developments in the field of biometric recognition. The use of multi-biometric recognition improves
the reliability and accuracy compared to the existing biometric system. An Automatic recognition system
uses biometric indicator and it produces high error rates [3]. Biometric vendors are used the Multi-biometric
system (e.g. fingerprint and finger vein by SAFRAN Morpho1) and it require large data sets (e.g. within the
Aadhaar project [4] by the Unique Identification Authority of India (UIDAI)). The multi-biometric system
provides multiple information about the same object and it the template of multi-biometric system requires
more security. Missing of any information from biometric template delete the person identity and creates
security problems. For instance permanency following from claiming subjects without assent [5], [6]
alternately remaking from claiming unique biometric qualities (e. G. Fingerprints [7] alternately iris textures
may turn into a sensible danger. That biometric format ought a chance to be secured by giving work to
protection Also integument about saved biometric information. Format protections schemes give provable
security alternately privacy, Furthermore useful recognition rates gotten are elusive, Indeed around little
datasets.

Int J Elec & Comp Eng ISSN: 2088-8708 
An Efficient Fingerprint Identification using Neural Network and BAT Algorithm (Subba Reddy Borra)
1195
Fingerprints are the unique characteristic of humanity. Fingerprint classification restores
computational complexity by reducing the number of candidate campaigns to be preceded [8]. Fingerprint
images are widely used in different systems, such as personal empathy, access control, Internet authentication
and the generation of encryption keys due to their durability, uniqueness and distinctive character. The
characteristic of the fingerprint system is affected by the image quality of the fingerprint [9]. The
identification of fingerprints, however, is a computer-based challenge especially for databases. An
operational indexing system provides a great help. Fingerprint classification classifies fingerprints into a set
of predefined sets, and then makes the identical task possible [10].
Fingerprint classification is a granular level apportioning of a large fingerprint database, where the
class of the input fingerprint is primarily persistent and therefore, a search is conducted inside the group of
fingerprints suitable to the similar class as the input fingerprint [11]. The authentication process is secured by
decreasing the number of comparisons that are essentially achieved. It is attained by isolating the fingerprint
database into a number of classes. Fingerprint recognition is prerequisite to be associated only to the
fingerprints in a single class of the database on the basis of its features. Fingerprint identification is broadly
used due to ease in the characteristic acquisition; the ten fingers are accessible for collection and their usage
and assortments of law implementation and immigration [12]. Fingerprint classification and acknowledgment
schemes are inspiring missions for hackers as it includes unique identification technique, incorporating Data
Mining methods such as Neural Networks and K Nearest Neighbor algorithms, the intelligence stage of the
finger print identification is enhanced which safeguard accuracy and protected recognition [13].
The rest of the document is organized as follows: Section 2 defines a concise description of the
latest research work; Section 3 refers to the processes that are part of our masterpiece method with surprising
demonstrations and mathematical formulations. While Section 4 elucidates the results of the experimentation,
Section 5 concludes the work.
2. RELATED WORK
Patil and Suralkar described a method of fingerprint classification scheme on the basis of ANN.
Fingerprint identification scheme usages prior organization of fingerprint with minutiae feature [14]. In
traditional methods performance of minutiae abstractions depend intensely on an augmentation algorithm.
Thus several fingerprints are composed, taking a long time to match and authenticate a specified fingerprint.
So as an alternative of classification with the minutiae [15] they projected a classification scheme that was on
the basis of individual features such as the singular point. Singular point detection was very robust and
reliable that overcomes the issue about rotation and translation.
Daramola et al. [16] projected a robust verification scheme on the basis of features abstracted from
human fingerprints and a pattern classifier known as Support Vector Machine (SVM). Efficient fingerprint
verification scheme was desired in numerous places for personal empathy to access physical facilities, data
etc. Three group of features are attached together and passed to the classifier. The fused feature was utilized
to train the scheme for operative verification of consumers fingerprint images.
Liu et al. [17] described a touch less multitier fingerprint capture scheme that acquires three
dissimilar aspects of fingerprint images at the similar time. This gadget might have been planned Eventually
Tom's perusing upgrading parameters in regards to those caught finger impression picture caliber and gadget
measure. This machine was intended by optimizing parameters concerning the captured fingerprint image
eminence and device size. A fingerprint mosaicking method was anticipated to splice together the captured
images of a finger to form an image with a greater valuable print arena. Optimization design of their device
was established by familiarizing the design process and associating with present touch less multitier
fingerprint acquirement devices.
Latent prints are habitually improved from offense scenes and are connected with accessible
databases of recognized fingerprints for recognizing criminals. However, current procedures to compare
latent prints to great databases of exemplar (rolled or plain) prints are predisposed to errors. This
recommended caution in creating conclusions about a suspect’s identity on the basis of a latent fingerprint
comparison. A cardinal of efforts accept been fabricated to statistically archetypal the account of a fingerprint
appraisal in authoritative a actual accept/reject accommodation or its apocalyptic value. These methods,
though, either accomplish unrealistic expectations about the archetypal or they abridgement simple
clarification.
Nagar et al. [18] have contended that the posterior probability of two fingerprints fitting to
dissimilar fingers provided their match score, mentioned to as the nonmatch probability (NMP), efficiently
detentions any associating indication of the comparison. NMP was calculated by state-of-the-art matchers
and was tranquil to understand. To integrate the effect of image quality, number of minutiae, and size of the

 ISSN: 2088-8708
Int J Elec & Comp Eng, Vol. 8, No. 2, April 2018 : 1194 – 1213
1196
dormant on NMP value, they calculated the NMP vs. match score plots distinctly for image pairs (latent and
exemplar prints) with dissimilar features.
Paulino et al. [19] projected a fingerprint matching algorithm that was particularly designed for
matching latents. Recognizing suspects on the basis of impressions of fingers lifted from crime scenes was a
routine process that is enormously significant to forensics and law implementation agencies. The projected
algorithm utilizes a robust alignment algorithm to align fingerprints and trials similarities within fingerprints
by taking into account both minutiae and orientation field data. To be consistent with the common practice in
latent matching the orientation field was reassembled from minutiae. Subsequently, the projected algorithm
depend only on manually marked minutiae, it could be effortlessly utilized in law enforcement solicitations.
Sankaran, et al. [20] focused on automatically segmenting latent fingerprints to distinguish between
ridge and non-ridge patterns. There were three real commitments about that method: (a) An machine Taking
in calculation to joining five diverse Classes of features for programmed idle finger impression segmentation,
(b) a feature selection procedure utilizing changed reduction plan for dissecting those impact of various
classification offers for idle finger impression segmentation, Furthermore (c) a novel SIVV built metric will
measure the impact of the segmentation algorithm without the prerequisite will perform those whole
matching transform. Those picture might have been decorated under nearby patches What's more saliency
based Characteristics alongside image, gradient, ridge, What's more personal satisfaction based features were
concentrated. Characteristic Choice might have been performed should ponder the commitment of the
Different classification features towards closer view edge example representational.
Aliakbarzadeh, et al. [21] indicated that the GC-FID chromatograms of saffron samples were
baseline corrected and aligned using asymmetric least squares (AsLS) and correlation optimized warping
(COW) methods, respectively. Then, those entirety advanced profiles from claiming preprocessed
chromatograms were normalized will internal standard, mean-centered, pareto-scaled Also At last
demonstrated finally perusing PLS-DA to arrange saffron specimens as stated by their development regions.
Afterwards, execution of distinctive variable determination techniques (i.e., rPLS, VIP, SR, sMC and loading
weights) for picking the the vast majority vital variables (i.e., maintenance occasion when points) for GC-
FID fingerprints, were compared As far as the model’s interpretability Also unoriginality.
Raid Al-Nima, S. S. Dlay et al [22] proposed human authentication method in which Finger Texture
(FT) patterns was used to make it efficient. To differentiate the fingers from the hand images, a robust and
automatic finger extraction method was used. Enhanced Local Line Binary Pattern (ELLBP) was used to
extract new features. The information embedded within the poorly imaged regions of the FTs a method is
suggested to salvage missing feature elements. Classification was done by performing Probabilistic Neural
Network (PNN).
Raid Al-Nima, S. S. Dlay et al [23] defined an approach which authenticates based on their finger
textures. Finger Texture (FT) features of the four finger images (index, middle, ring and little) are extracted
from a low resolution contactless hand image. To enhance the FTs, new Image Feature Enhancement (IFE)
was used method to enhance the FTs. The resulting feature image is segmented and a Probabilistic Neural
Network (PNN) is employed to classify intelligently for recognition.
Fingerprint matching is a critical issue to recognizing fingerprints What's more assumes a way part
in the fingerprint distinguishment systems. However, performing fingerprint ID number in an expansive
database might a chance to be an wasteful errand because of those absence of versatility Also secondary
registering times from claiming fingerprint matching calculations. Fingerprint indexing may be an magic
system to programmed fingerprint ID number frameworks which permit us to decrease the number of
candidates, the look space, and the occurrences of false acknowledgement clinched alongside substantial
databases. Javad Khodadoust and Ali Mohammad Khodadoust, [24] proposed an efficient indexing algorithm
using minutia pairs and convex core point which employed k-means clustering and candidate list reduction
criteria to increase the recognition performance. Their proposal could successfully reduce the search space
and number of candidates for fingerprint matching, and thus achieved higher matching scores and
considerably improved the system repossession performance.
3. NEURAL NETWORK AND BAT ALGORITHM BASED FINGERPRINT CLASSIFICATION
Fingerprint image classification has become an interesting topic in recent years because of the
unique nature of fingerprints in the authentication process. The fingerprint authentication is regarded as one
of the most secure ways of protecting the user information’s due to its uniqueness and durability. Due to the
widespread usage, there also exists some disadvantage like creating fake fingerprints inorder to access the
information. Hence the classification of fingerprint has become a much needed process while using
fingerprint as an authentication technique. We have designed an efficient technique for fingerprint
classification using soft computing techniques. The proposed classification system can be divided into three

1197
stages. The first stage is the preprocessing stage where the fingerprint images are subjected to noise removal
using the median filter. After noise removal histogram equalization is performed for enhancing the images.
The second stage is the feature extraction stage. Here various image features like Area, holo entropy, SURF
and SIFT features are extracted. The final stage in the proposed system is classification. We have utilized
hybrid neural network for classification of fingerprint as fake or original.
3.1. Problem definition and Objectives
Fingerprint technology is a biometric method that is used to recognize persons on the basis of their
physical traits. Fingerprint based authentication systems are becoming increasingly common these days.
However, due to the excessive use of fingerprint security systems, they have become a target of attacks. The
reputation of the fingerprint image regulates the durability of a fingerprint authentication scheme. We have
designed an efficient technique for fingerprint classification using soft computing techniques.
Objectives:
a. To protect biometric template with high security.
b. To avoid cross-comparison of biometric templates in databases.
c. To establish soft biometrics idea.
d. To provide revocability to biometric templates.
e. To impart revocability to biometric templates.
f. To permit diversity to biometric templates.
g. To make sure that the authorized user is recognized perfectly.
3.2. Fingerprint Classification
The neural network is incorporated with BAT algorithm for optimizing the weight factor. The flow
diagram of our proposed system is shown in Figure 1.
Figure 1. Flow diagram for proposed method
3.3. Noise Removal using Median Filter
Noise removal is a significant stage in image processing. In our projected technique we apply a
median filter for the noise removal. The median filter is frequently employed to overcast worth images due to
Fingerprint image Median filtering for
noise removal
Histogram
equalization for
enhancing images
Hybrid Neural network BAT Algorithm
Fingerprint image Classification
Original Fingerprint Fake Fingerprint

 ISSN: 2088-8708
1198
its belongings of control protective flattening. In the median filtering operation, the pixel values in the
vicinity window are graded according to intensity, and the median value turn out to be the output for the pixel
under evaluation. The succeeding stages were accustomed to eliminate the noise from images,
In median filtering, the adjacent pixels are graded in line with illumination and the average value
turn out to be the new central pixel value. Median filter ensures the brilliant work of rejecting particular
forms of noise, in the actual, “shot” or impulse noise in which some individual pixels have higher values. The
common countenance for the median filter is assumed as per Equation (1),
),......min()......,(
11
121  

m
l
lm
m
l
lmmf xxxxxxxO (1)
Where,
mfO - Median Filtered output
mxxx ......, 21 - Number of pixels under evaluation
m - Number of pixels
Here, m represents the total number of pixels in the image and mxxx ......, 21 represents the pixels
from first to m. With Equation (1), the median filtering is executed to eliminate the noise from the developed
image. Once the noise removal stage is completed, the next stage is image enhancement where we utilize
histogram equalization for enhancing the image.
3.4. Image Enhancement using Histogram Equalization
The histogram equalization represents a computer image processing method effectively employed to
enhance the contrast in images. Let the original input be I and the equalized image of the histogram be hI .
The underlying motive behind the novel technique is involved in creating an enriched image, equipped with
enhanced visual excellence far superior to I .However, it is easy to design a renovation function which is
well-geared to automatically realize the relative mapping impact, only in accordance with the data existing in
the histogram of the input image.
Each bin representing the constraint of HE of a histogram in a gray scale image indicates the
number of pixels with the identical gray value in the image. The probability distribution function (pdf) of
input image and improved image are labeled as IPD , and xIPD and related cumulative distribution
functions (cdf) are ICD , and xICD correspondingly. Further the intensity of the original input image is
defined as g and the adapted intensity of the output image as 0g . As the output probability distribution
function has to homogenous, i.e. opPD =
LI
1
, where LI represents the number of intensity levels. The
superlative output cumulative distribution functions is illustrated by means of the following Equation (2).
L
g
r
op
G
g
gCD
1
)( 0
0
0

 
For 1,....2,1,00  LGg (2)
As )(gCDI = )(gCDop subsequent to the histogram equalization, it is possible to attain the
output intensity by means of the above Equation (2). Here, LG specifies the gray level image intensities in
the range of 0–255. In this regard, the flat, bell-shaped, and curved histograms constitute certain preferred
histogram shapes for the image tiles. It incredibly improves the contrast of images by appropriately varying
the intensity values, to usher in the output image whose histogram harmonizes with a homogeneous
histogram. The transformation X for minimization is represented by means of the following Equation (3).
rI CD-(X(r))CD (3)

1199
where ICD the cumulative histogram of X, and rCD represents the cumulative total of histograms for all
intensities r . This minimization is subjected to various parameters such as that X has to be monotonic in
nature and  rXCDI must not exceed ICD by more than 50% of the distance between the histogram
counts at g . With an eye on cutting back the flaws in Equation (3) and adapting X as a monotonic
function, a homogeneous HE input is furnished to Equation (3). The corresponding transformation is
efficiently employed to map the gray levels in X to their new values with superior monotonic function.
3.5. Feature Extraction
Once the image is enhanced to improve the quality, the image is subjected to feature extraction
where different extensive features like Area, holo entropy, SURF and SIFT features are extracted. The
features extraction is a main stage in image processing as it provides the parameters for evaluating the image
for classification purpose. Each feature extracted are explained below,
3.5.1. Area
The area is extracted in order to find out the exact position of the fingerprint in the image. The area
is estimated using the expression given below,
whAArea , (4)
Where,
h - indicates the image height.
w- denotes the image width.
3.5.2. Holo Entropy
Holo entropy is an arithmetical measure of unpredictability that can be used to distinguish the
texture of the input image. The holo entropy basically depends on the histogram value in the image.
The holo entropy can be measured in terms of summation of entropies on all the attributes in the
particular section of image.
The holo entropy is defined as the ration of sum of entropy to that of total correlation which can also
be represented in terms of summation of entropies which is given as per Equation (5) below,


N
k
kmH pXIX
1
)()( (5)
Where,
)( mH IX - Holo entropy for the image mI
)( kpX - Entropy of the image pixel.
3.5.3. SIFT and SURF Feature
3.5.3.1. SIFT Feature
The scale invariant feature transform (SIFT) is a powerful feature that is used in image retrieval
process. The approach is performed by detecting the scale space extremea in the image where we identify the
locations and scales in the image. The scale space of an image ),( nmZ is represented by a function
),,( baFs which is given by,
),(),,(),,( nmZnmGnmF fs   (6)
where,
),,( nmGf - Gaussian function which is given by

 ISSN: 2088-8708
1200
2
22
2
)(
2
2
1
),,( 


nm
f enmG


 (7)
The sift feature vector are thus calculated using the above expression and these feature values are then used
for further processing.
3.5.3.2 SURF Feature
The proposed facial recognition technique uses a robust feature extraction method (SURF) to extract
the characteristics of the training and test image. The extraction method SURF is a method of extraction of
insignificant scaling and rotational properties, which is faster than the commonly used extraction method.
Transformation of invariant characteristic of the scale. SURF focuses on the invariant detectors on the scale
and on the rotation in the plane and the descriptors of an image. The essential image is calculated from the
image and calculates the sum of the pixel intensities in the integrated image using the following equation
below,
 





pa
a
qb
b
bahqph
0 0
),(),( (8)
The hessian matrix is used for determining the intensity point in SURF feature extraction. The
matrix is calculated using the expression below,









),(),(
),(),(
),(
,,
,,



gIgI
gIgI
gH
hhhg
hggg
m (9)
Where ),(, gI gg to ),(, gI hh - Convolution of the Gaussian second order derivative.
The locations in the image where the determinant of Hessian matrix is maximum are detected. Pixel
intensities are high where the determinant of Hessian matrix is maximum, so determinant of Hessian matrix
gives the maximum intensity points in an image. The features of these maximum intensity points are
extracted to implement the proposed Fingerprint image classification.
Once the features are extracted, these feature vector values [A, XH, Fs, h ] are applied to the classifier for
classification of images which is explained in below section.
3.6. Fingerprint Classification Using Hybrid Neural Network
The Hybrid Neural Network is used to discover the unique classification of fingerprints and is
prepared using the values of the components extracted from each image. The Hybrid Neural Network is well
trained by extorted characteristics. The Hybrid Neural Network has four input units, n hidden units, and one
output unit. Here n represents the number of hidden layer neurons. In our proposed method, the number of
hidden neurons will be approx. 20. The contribution of the neural system is the element vector we have
extricated from the images. The system is prepared under an extensive arrangement of various images from
info database so as to empower them to successfully order the correct unique finger print image in the testing
stage. The Hybrid Neural Network works making utilization of two stages, one is the training phase and the
other is the testing phase.
a. Learning Algorithm – Back Propagation Algorithm
The Back Propagation Algorithm is efficiently exploited as the Learning algorithm in the Feed
Forward Neural System. The Back Propagation Algorithm essentially signifies a managed Learning policy
and supplementary it distinguishes the collapse of delta regulation. For the principle of execution
compilation, it necessitates of a dataset of the fundamental efficiency for diverse inputs. Typically, the Back
Propagation Algorithm is the ultimate one for the Feed-Forward Networks and the Learning algorithm
necessitate that the service principle engaged by means of neurons must be varied [25].
b. Back propagation Algorithm Steps for FFBNN
The weight for the neurons of hidden layer and the output layer are premeditated erratically decides
the heaviness. Though, the input layer acquires the steady weight.
The predictable Bias task and the establishment task are estimated through Equations (10) and (11) for the
FFBNN.

1201
The Back Propagation fault is estimated for each one node and subsequently, the weights are restructured as
per the subsequent Equation (11).
)()()( '''
nnn
www 
(10)
The weight
)( '
n
w is adapted as per Equation (11) shown below.
)(
)()(
.. ''
BP
nn
EXw 
(11)
Where,
 - Learning Rate, which is habitually in the range of 0.2 to 0.5.
)(BP
E - BP Error.
The process is constant by the aid of steps signify in Equations (10) and (11), till the BP fault is
condensed to the least i.e. .1.0)(
BP
E
On accomplishing the least value, the FFBNN become known suitably appropriate for the
transmission segment.
Accordingly, the FFBNN classifier is successfully practiced and the organization standards are
experienced through utilizing the characteristic. The categorization of fault is implemented which categorize
the complete fault engender from the classifier.
The Hybrid Neural Network works by using two stages, one is the training phase and the other is the
testing phase.
3.6.1. Training Phase
In the training phase, the input image is extracted from features and this feature vector is provided as
the input to the neural network. Initially, the nodes receive random weights. As the output is already known
in the training phase, the output obtained from the neural network is compared with the original and the
weights vary to reduce the error. This process is carried out for large data in order to produce a stable system
with assigned weights in the nodes. In our methodology we use a multilayer feed forward neural network.
The structure is represented in Figure 2.
Figure 2. Structure of artificial neural network
3.6.2. Testing Phase
In the test phase, the input image is fed to the trained neural network that has particular weights in
the nodes and the output is calculated to classify the images according to the trained data set. In the ordinary

 ISSN: 2088-8708
1202
neural network, the process will stop after the test. In the proposed modified neural network, for the test
process, we have incorporated the optimization algorithm to optimize the weight used for the tests. In our
proposed method, the weights are optimized with the help of the BAT Algorithm. By incorporating the
optimization process, the accuracy of the classification will be improved there by providing a better
classification of the images. The structure of the artificial neural network is illustrated in Figure 2.
3.6.3. Bat Algorithm for Optimizing Weights in ANN
The bat algorithm is a metaheuristic algorithm, excited by the behavior of echolocation of microbats
[26-34]. The bat algorithm (BA) is used to optimize the weight of hidden layer neurons in the neural network.
The BA that exploits this echolocation function depends on some of the important parameters, such
as frequency, velocity, pulse rate and loudness. The BA changes to the optimal solution when updating the
current position with the velocity of the most suitable solutions. Now, the pulse emission rate, as well as the
loudness, is also efficient in each iteration. In the demand to achieve the BA some of the expectations were
predefined depending on the characteristic characteristics of the bats.
Assumption:
Several of the assumptions must be made within the Bat (BA) Algorithm. Expectations are irregular
here,
a. All bats are able to distinguish between background and Prey.
b. All bats use echolocation property to detect distance.
c. All bats fly unsystematically with velocity iv in position ix and releases pulses of sound with
frequency if , fluctuating wavelength  and loudness il .
d. Frequency (or) wavelength varies established in the vicinity of the target particle.
e. Furthermore, the pulse emission rate can also be varied between the range of 0 and 1 based on the
target location.
f. Loudness il congregates among the maximum loudness maxl to constant minimum loudness minl
The following flow illustration for the bat algorithm used to optimize the whole value is assumed by
the Figure 3.
Step 1: Input micro-bats ( iB ) population is randomly generated. According to our proposed approach, the
weights of neurons are considered as the micro-bats. Each micro-bat has the velocity vector )( iv and position
vector )( ix , which is described by the following Equation (12). Initially, the values of these credentials are
assigned randomly to a particular range.















bn
mnmn
b
mm
b
mm
bn
nn
bb
bn
nn
bb
i
xvxvxv
xvxvxv
xvxvxv
B
),(),(),(
),(),(),(
),(),(),(
2
22
1
11
22
2
2222
2
2221
11
2
1212
1
1111




(12)
Step 2: To allocate the echolocation parameters, the micro-bat populations are included with the
echolocation parameters like frequency )( if , pulse rate )( ipr , and the loudness parameters )( il . These
parameters are non-negative real values with the following ranges.
maxmin fff i  , maxmin prprpr i  , maxmin lll i  (13)
Here, we allocate the frequency range 0min f and 1max f , the pulse rate minimum value
5.0min pr is and the loudness maximum value is 1max l . The remaining values are determined by the
subsequent Equation (14).
sec
min
1
n
l  (14)

1203
1
1
1max 
dn
pr (15)
Where, secn is the number of sections in the discrete set used for sizing the design variable and secn is the
number of discrete design variables.
Step 3: Calculate the objective function of the initial populations; the required fitness function is described
by the following Equation (16).
MSEFi min (16)
According to Equation (16), the frequency of each class label is used to define fitness value for each
microbat. The fitness function of the micro bats is determined based on the mean square error (MSE). When
the obtained MSE of a micro bat is found low, then the micro bat is ranked as a best micro bat.
Step 4: Store the current population and augment the iteration count as t+1, i.e., iteration t = t+1.
Step 5: The current population of rules is randomly updated based on the frequency and the velocity.
Initially, the frequency can be evaluated, which is described by the following Equation (17).
i
t
i uffff )( minmaxmin  (17)
Where, iu is the random number of values, which is selected from 0 to1, then the frequency is applied to
the velocity equation, which can be described by the following Equation (18).
])([ 11 t
i
t
i
t
i
t
i fxxvroundv 

 (18)
t
i
t
i
t
i vxx  1
(19)
Where,
t
iv and
1t
iv are the velocity vectors of the micro-bats at the time steps t and,
t
ix and
1t
ix are the
position vectors of the micro-bats at time steps t and, x is the current global best solution. Here after
performing the local search in the randomly selected population, this is described by the following
Equation (20). A solution is selected among current best solutions and then random walk is introduced to
obtain new solution
t
avgjioldnew lxx , (20)
Where, ji, is a random number between 1 and1,
t
avgl is the average value of loudness at time step.
Step 6: Find the fitness of the new micro-bats population using the Equation (16). After evaluation, the
micro-bats echolocation parameters are updated for better moving of the micro-bats, which can be described
by the following Equation (21).
)]exp(1[. max
11
tprprandlal t
i
t
i
t
i  
(21)
Where,
1t
il and
t
il are the updated and previous values of the loudness
1t
pr is the pulse rate of the micro-
bats in the time step, a and are the adaptation parameters of the loudness and pulse rate.
Step 7: To find the best micro-bats, which satisfies the objective function.
Step 8: The steps 4 to 7 is continued until the termination criteria are attained.
Here the input will be the weights of Neuron, based on the fitness; the optimal weights of a neuron
are selected. The optimal weights will help the ANN to classify the fingerprints more accurately. The fitness
of BA algorithm is the minimization of MSE. For a neural network, we have to minimize the error, to get the
accurate classification.

 ISSN: 2088-8708
1204
Figure 3. Bat Algorithm
By utilizing the BAT algorithm as shown above, the weights are assigned to the NN which makes
the classification of images more accurate and improved. The Hybrid Neural Network delivers better
accuracy in terms of classifying the exact image because of incorporating the optimization process. Various
input images are applied to our proposed system and using the extensive features extracted for each image,
the classification is done with the aid of proposed Hybrid Neural Network and the results with improved
classification accuracy are evaluated
4. RESULTS AND DISCUSSION
4.1. Database Description
The experiment is done in MATLAB (2015a) by applying the proposed approach and Image
Processing Toolbox was used to produce the improved finger print image. The propped approach is validated
using FVC2000 dataset. The original and fake fingerprint images are collected through Fingerprint
Verification Competition or FVC2000 [30] and the samples are drawn from SFinGE. Different sensors are
applied on FVC2000 to collect four databases from FVC2000 database. Low cost Optical sensor was used to
collect images for DB1. Low cost Capacitive Optical sensor used to collect images for DB2. DB3 is collated
using a fairly sizeable quality of optical sensors. At final, databases DB4 is synthetically generated using
SFinGE. Here the total number of images used is nearly 300. In this, we have taken 90-10 ratio for training
No
Yes
No
Yes
Calculate frequency and velocity
Initialize pulse rate & loudness
Evaluate fitness
Generate new solution
Find best solution
Calculate fitness
Max iteration
reached
Rand >pulse
rate
Position bat and find best solution
Start
Stop

1205
and testing. The results will be analyzed by varying the training and testing ratio such as 80-20 and 70-30.
The data base specifications are given in Table 1 and the sample images are shown from Figure 3 to
Figure10.
Table 1. The four FVC2000 databases
Database names Sensor type Image Size
DB1 Low-cost Optical 388 X 300
DB2 Low-cost Capacitive Optical sensor 256 X 364
DB3 Optical sensor 448 X 478
DB4 Synthetic generator 240X 320
Figure 3. DB1 sample images; each row shows different images of fingerprints of the same finger
Figure 4. Fingerprint images of DB1; all samples are from different fingers and are prepared by quality (top
left: high quality, bottom right: low quality)
Figure 5. DB2 sample fingerprint images; each row shows different images of fingerprints of the same finger

 ISSN: 2088-8708
1206
Figure 6. DB2 fingerprint images; all samples are from different fingers and are prepared by quality (top left:
high quality, bottom right: low quality)
Figure 7. Fingerprint images of DB3; each row shows different images of fingerprints of the same finger
Figure 8. Fingerprint images of DB3; all samples are from different fingers and are ordered by quality (Top
left: high quality, bottom right: low quality)

1207
Figure 9. DB4 sample images; each row shows different images of fingerprints of the same finger
Figure 10. DB4 images; all samples are from different fingers and are ordered by quality (top left: high
quality, bottom right: low quality)
S.No Input images Enhanced image Classified Results
1.
2.
3.
4.
5.
Figure 11. Processed output of fingerprint image classification

 ISSN: 2088-8708
1208
Figure 11 show the specified the processed output for the input fingerprint images.
As publicized in above figure, the input image is substance to noise reduction and then improved by
the Histogram equalization technique. The improved images are then classified on the basis of the input
query image using Hybrid Neural Network. The categorized images are then stowed for empathy. The
presentations of proposed technique in organization of exact images are then assessed and are associated with
that of available neural network.
4.2. Performance Evaluation
The performance of the proposed approach is validated by using metrics namely Precision,
sensitivity, specificity, accuracy and F-Measure. These metrics are assessed for different training testing
percentages and are tabularized. Similar metrics for the existing methods are also evaluated and tabulated for
comparing with proposed method. In our proposed scheme we have utilized neural network which is one of
the existing method for classification. The performance metrics are defined in Table 2. Table 3 show the
Performance assessment of Proposed Method (NN+BAT)
Table 2. Performance metrics
True Positive(tp) The number of images recognized as correct, which are actually correct
False Positive(fp) The number of images recognized as correct, which are actually out of classification or wrong.
True Negative(tn) The number of images recognized as wrong or out of classification , which are actually wrong or out of
classification
False Negative(fn) The number of images recognized as wrong or out of classification, which are actually correct.
Precision tp
tp fp
Recall/sensitivity tp
tp fn
Specificity tn
tn fp
Accuracy tn tp
tp tn fp fn

  
F-Measure
2
recall precision
recall precision
 
 
 
a. Precision shows the class agreement of the data labels with the positive labels given by the
classifier.
b. Sensitivity shows the effectiveness of a classifier to identify the positive labels.
c. Specificity shows how effectively a classifier identifies the negative labels.
d. Accuracy shows the overall effectiveness of a classifier.
e. F-measure shows the relation between data’s positive labels and those given by a classifier.
Table 3. Performance assessment of Proposed Method (NN+BAT)
Training –Testing Percentage 90-10 80-20 70-30
Accuracy 0.987513 0.978514 0.958265
Sensitivity 0.98479 0.986572 0.995466
Specificity 0.95625 0.87964 0.987463
Precision 0.9756 0.9652 0.9453
F-Measure 0.9375 0.9084 0.9532

1209
The Figure 12 shows the metrics values like Precision, sensitivity, specificity, accuracy and
F-Measure accomplished by the anticipated methods for various training and testing quantities.
Figure 12. Evaluation of Performance Metris for different training and testing ratios
The Table 4 specified below exemplifies the metrics values like Precision, sensitivity, specificity,
accuracy and F-Measure accomplished by the anticipated and predominant methods. From the values
accomplished it is apparent that our anticipated system has outdone the existing method
Table 4. Performance assessment of proposed and prevailing methods
Training –
Testing
Percentage
Proposed Method (NN+BAT)
Neural Network with PSO
algorithm
Neural network Without Optimization
90-10 80-20 70-30 90-10 80-20 70-30 90-10 80-20 70-30
Accuracy 0.987513 0.978514 0.958265 0.9738 0.9684 0.9465 0.9625 0.958962 0.930789
Sensitivity 0.98479 0.986572 0.995466 0.98 0.9532 0.94566 0.975 0.92479 0.925896
Specificity 0.95625 0.87964 0.987463 0.9 0.8054 0.96462 0.75 0.75478 0.9563
Precision 0.9756 0.9652 0.9453 0.9723 0.9525 0.9385 0.9638 0.935 0.925
F-Measure 0.9375 0.9084 0.9532 0.9274 0.9045 0.9325 0.9075 0.8875 0.9025
Hence the above performance evaluation shows Precision, sensitivity, specificity, accuracy and F-Measure
over Neural network without optimisation , Neural network with PSO algorithm and Neural Network with
BAT algorithm. Therefore it is proved that our proposed NN with BAT algorithm working efficiently
comparing to the existing methodolgies mentioned above. The below Figure 13 shows the comparison of all
the above paramters for proposed and exisitng methods.
In [35] order to improve the limitations of existing fingerprint image enhancement methods an
efficient technique is proposed for image processing with low image quality fingerprint images. The
methodology is divided into three modules. In the first module, the fingerprint image is subjected to
denoising process where Wave atom transform is utilized. After the completion of this process the image
enhancement is performed with the help of optimization algorithm. In ths approach, a Modified cuckoo
search (MCS) algorithm is used as an optimizer. This helps to come across for the best gray level distribution
that maximizes the objective function.
In [36 ] to enhance the restrictions of prevailing fingerprint image augmentation approaches authors
projected an effectual method to pact with small quality fingerprint images. The projected methodology is
alienated into three modules. Primarily, the fingerprint image is endangered to denoising procedure where
Wave atom transform is used. Once this procedure is accomplished the image augmentation is achieved for
improving the classification rate. The morphological operation is used in this projected technique inorder to
augment the image. The morphological operators such as dilation and area opening are used here for

 ISSN: 2088-8708
1210
improvement. The final step in this is the ordering of fingerprint image. In this auhors used Adaptive genetic
neural network (AGNN) intended for classification of images.
Figure. 13 Graphical representation of Performance metrics for proposed and prevailing methods
Table 5. Performance assessment of proposed and existing methods
The proposed method uses the neural network and BAT algorithm in place of wave atom
transformation to by minimizing the time improve the performance of detection by minimizing the detection
time which is shown in Figure 15. Comparative analysis of performance metrics are shown in Table 5 and
Figure 14.
Figure14. Graphical representation of Performance metrics for proposed and existing methods
Training –
Testing
Percentage
(%)
Proposed Method (NN+BAT)
Wave Atom Transformations+ AGNN
Method
Wave Atom Transformations+
Modified cuckoo search (MCS)
algorithm
90-10 80-20 70-30 90-10 80-20 70-30 90-10 80-20 70-30
Accuracy 0.987513 0.978514 0.958265 0.978125 0.964785 0.948265 0.9730 0.9545 0.9248
Sensitivity 0.98479 0.986572 0.995466 1 0.986572 0.985466 0.9435 0.9748 0.9785
Specificity 0.95625 0.87964 0.987463 0.95625 0.938796 0.925875 0.9435 0.9280 0.9175
Precision 0.9756 0.9652 0.9453 0.9756 0.9652 0.9453 0.9650 0.9450 0.9385
F-Measure 0.9375 0.9084 0.9532 0.9375 0.9084 0.9532 0.9250 0.90 0.9450

1211
Figure 15. Detection time comparision of Proposed method with existing methods (X-Axis no.of images and
Y-axis Detection timein Milli seconds)
5. CONCLUSION AND FEATURE WORK
In this effort, in order to overawe the fingerprint images classification issues in dissimilar available
plan, an effective fingerprint image classification technique has been projected. The biometric fingerprint
classification system is used in criminal identification and human authentication process in real time. In order
to ensure the security and authorization, we need to identify the original fingerprint. For this purpose, we are
developing a fingerprint classification system. The projected method uses median filter for noise removal,
Histogram equalization for image augmentation and Hybrid neural network for image classification. The
projected approach aided in classifying the images into original or fake meanwhile the images are improved
and utilization of hybrid neural network. Numerous parameters such as accuracy, sensitivity,
specificity,Precision and F-Measure are assessed by the projected method and are then associated with the
available neural network. From the solution attained it is vibrant that our projected scheme of fingerprint
image classification overtakes the available technique with improved classification accuracy. To improve the
template security of biometric systems in future work are given below
To improve the performance of individual traits and to raise the performance of the overall system
lot of matching algorithms are discussed. For real time application any kind of Biometric System can be
used. It needs various sensors to capture the data and fetch the same process in the real time application for
training and testing phase. There should not be noise, illumination changes and background variations in the
captured data. The less memory consumption and computational effort can be improved in further proposed
scheme.
REFERENCES
[1] C. Rathgeb and A. Uhl, “A survey on biometric cryptosystems and cancelable biometrics,” URASIP Journal on Information
Security 2011, 2011:3.
[2] A. Ross and A. K. Jain, “Information fusion in biometrics,” Pattern Recognition Letters, 24:2115–2125, 2003.
[3] Unique Identification Authority of India. Aadhaar, 2012. Retrieved May, 2012.
[4] A. Nagar, K. Nandakumar, and A. K. Jain, “Multibiometric cryptosystems based on feature level fusion,” IEEE Transactions on
Information Forensics and Security, 7(1):255–268, 2012.
[5] N. K. Ratha, J. H. Connell, and R. M. Bolle, “Enhancing security and privacy in biometrics-based authentication systems,” IBM
Systems Journal, 40:614–634, 2001.
[6] R. Cappelli, A. Lumini, D. Maio, and D. Maltoni, “Fingerprint image reconstruction from standard templates,” IEEE Transactions
on Pattern Analysis and Machine Intelligence, 29(9):1489–1503, 2007.
[7] S. Venugopalan and M. Savvides, “How to generate spoofed irises from an iris code template,” Trans. Information Forensics and
Security, 6:385–395, 2011.
[8] Ching-Tang Hsieh, Shys-Rong Shyu and Kuo-Ming Hung, "An Effective Method for Fingerprint Classification”, Tamkang
Journal of Science and Engineering, Vol.12, No.2, pp. 169-182, 2009.
[9] M.S. Altarawneh, W.L.Woo and S.S Dlay, "Objective Fingerprint Image Quality Assessment Using Gabor Spectrum Approach",
15th International Conference on Digital Signal Processing, Cardiff University Wales, UK, pp. 248-251, 2007.
[10] Qinzhi Zhang, Kai Huang and Hong Yan, "Fingerprint Classification Based on Extraction and Analysis of Singularities and
Pseudoridges", Pattern Recognition, Vol. 37, No.11, 2004.
[11] Sarat C. Dass and Anil K. Jain, "Fingerprint Classification Using Orientation Field Flow Curves", Fourth Indian Conference on
Computer Vision, Graphics & Image Processing, Kolkata ,India, pp. 650-655 ,2004.
[12] Suman Sahu, A. Prabhakar Rao and Saurabh Tarun Mishra, "Fingerprints based Gender Classification using Adaptive Neuro
Fuzzy Inference System", IEEE conference, pp. 1218-1222, 2015.

 ISSN: 2088-8708
1212
[13] K.Umamaheswari, S. Sumathi, S.N. Sivanandam and K.K.N. Anburajan, "Efficient Finger Print Image Classifi'cation and
Recognition using Neural Network Data Mining", IEEE - ICSCN, pp.426-432,2007.
[14] Dimple Parekh and Rekha Vig, "Survey on Parameters of Fingerprint Classification Methods Based On Algorithmic Flow",
International Journal of Computer Science & Engineering Survey (IJCSES) Vol.2, No.3, 2011.
[15] S R Patil and S R Suralkar, "Neural Network based Fingerprint Classification", International Journal of Science and Research
(IJSR), Vol.2, No.1, 2013.
[16] S. Adebayo Daramola, Tola Sokunbi and A.U Adoghe, "Fingerprint Verification System Using Support Vector Machine",
International Journal on Computer Science and Engineering, Vol. 5, No. 07, 2013.
[17] Feng Liu, David Zhang,Changjiang Song, and Guangming Lu, "Touchless Multiview Fingerprint Acquisition and Mosaicking",
IEEE Transactions On Instrumentation And Measurement, Vol. 62, No. 9,2013.
[18] Abhishek Nagar, Heeseung Choi and Anil K. Jain, "Evidential Value of Automated Latent Fingerprint Comparison: An Empirical
Approach”, IEEE Transactions On Information Forensics And Security, Vol. 7, No. 6, 2012.
[19] Alessandra A. Paulino, Jianjiang Feng, and Anil K. Jain, "Latent Fingerprint Matching Using Descriptor-Based Hough
Transform", IEEE Transactions On Information Forensics And Security, Vol. 8, No. 1,2013.
[20] Sankaran, Anush, Aayush Jain, Tarun Vashisth, Mayank Vatsa and Richa Singh, "Adaptive latent fingerprint segmentation using
feature selection and random decision forest classification", Information Fusion, Vol.34, pp.1-15, 2016.
[21] Aliakbarzadeh, Ghazaleh, Hadi Parastar, and Hassan Sereshti, "Classification of gas chromatographic fingerprints of saffron using
partial least squares discriminant analysis together with different variable selection methods", Chemometrics and Intelligent
Laboratory Systems, Vol.158, pp.165-173, 2016.
[22] Raid Al-Nima, S. S. Dlay, Al-Sumaidaee, W. L. Woo and J. A. Chambers, “Robust Feature Extraction and Salvage Schemes for
Finger Texture Based Biometrics,” IET Biometrics, vol. 6, no. 2, pp. 43-52, 2016
[23] R. Al-Nima, S.S. Dlay, W.L. Woo and J.A. Chambers, “Human authentication with finger textures based on image feature
enhancement” in Proc. of 2nd
IET International Conference on Intelligent Signal Processing (ISP), 2015, London, UK, pp. 1-6.
[24] Javad Khodadoust and Ali Mohammad Khodadoust, “Fingerprint indexing based on minutiae pairs and convex core point”,
Pattern Recognition, pp.1-23, 2016.
[25] Shamik Tiwari,”A Pattern Classification Based Approach for Blur Classification”, Indonesian Journal of Electrical Engineering
and Informatics (IJEEI), Vol.5, No.2, pp.162-173, 2017.
[26] Mo Yuanbin, Zhao Xinquan and Xiang Shujian ,”Local Memory Search Bat Algorithm for Grey Economic Dynamic System”,
Indonesian Journal of Electrical Engineering and Computer Science(IJEECS), Vol. 11, No. 9, pp.4925-4934, 2013.
[27] Djossou Adeyemi Amon,”A Modified Bat Algorithm for power loss Reduction in Electrical Distribution System”, Indonesian
Journal of Electrical Engineering and Computer Science (IJEECS), Vol. 14, No. 1, pp.55-61, 2015.
[28] Bestha Mallikarjuna ,K.Hemanth Reddy and O Hemakeshavulu, “Economic local Dispatch with valve-point Result Employing a
Binary Bat Formula”, International Journal of Electrical and Computer Engineering (IJECE), Vol. 4, No. 1, pp.101-107,2014.
[29] Xin-She Yang, "A New Metaheuristic Bat-Inspired Algorithm", Nature Inspired Cooperative Strategies for Optimization (NICSO
2010), pp 65-74.
[30] Xin-She Yang, "Bat Algorithm for Multi-objective Optimization", Optimization and control, Vol.1, pp.1-12, Mar 2012.
[31] Xin-She Yang, "Bat algorithm for multi-objective optimisation", Journal International Journal of Bio-Inspired Computation,
Vol. 3, No. 5, pp. 267-274, 2011.
[32] Maio, Dario, et al. "FVC2000: Fingerprint verification competition," Pattern Analysis and Machine Intelligence, IEEE
Transactions on 24.3 (2002): 402-412.
[33] R. González, D. A. Pelta, C. Cruz, G. Terrazas, and N. Krasnogor, Springer-Verlag, Berlin Heidelberg (2010), pp. 65-74.
[34] P.-W. Tsai, J.-S. Pan, B.-Y. Liao, M.-J. Tsai, and V. Istanda, “Bat Algorithm Inspired Algorithm for Solving Numerical
Optimization Problems”, Applied Mechanics and Materials Vol. 148-149 (2012) pp 134-137.
[35] Subba Reddy Borra, G. Jagadeeswar Reddy, and E.Sreenivasa Reddy, “An Efficient Fingerprint Enhancement Technique Using
Wave Atom Transform and MCS Algorithm “, IMCIP-2016, Computer Procedia, Vol. 89, pp.785-793, 2016.
[36] Subba ReddyBorra, G.Jagadeeswar Reddy, E.Sreenivasa Reddy, "Classification of Fingerprint Images with the aid of
Morphological Operation and AGNN Classifier", Applied Computing and Informatics, 2017.
BIOGRAPHIES OF AUTHORS
Subba Reddy Borra Obtained his Bachelor of Technology (Computer Science and Engineering)
from BAPATLA ENGINEERING COLLEGE, from 1998 to 2002, and he received M.Tech
(NEURAL NETWORKS)(CSE) , JNTK KAKINADA from 2009 to 2011 and he is doing Ph.D ,
JNTUH HYDERABAD . He is currently working as an Assoc. Professor in Department of
Computer Science and Engineering at Dr Samuel George Institute of Engineering and
Technology, Markapur-523316, India. His areas of interest are Digital Image Processing,
Machine Learning, pattern Recognition. Date of Birth: 05-06-1978. Year of registration: 2012
Dr. G. Jagadeeswar Reddy received B.Tech(Electronics&Communication Engg) from S.V.
University in May 1995,M.Tech(Industrial Electronics) from Mysore University ,S.J.C.E, Mysore
in 1998, Ph. D from JNTU HYDERABAD IN 2011. He has 17 years of teaching experience and
Currently he is working as Principal, Vaagdevi Intitute of Technology and Science ,Proddatur
,AP,India .He has Got Best paper award for the paper “Lossy Compression and
CurveletThresholding for Image Denoising” presented at International Conference on Electronic
Design conducted at Malaysia in the year 2008.He Got First Prize in the test in DSP short term
course conducted by Sri Nidhi Institute of Technology, Hyderabad in the year 2002.His areas of

1213
interest are Digital Image Processing,Digital Signal Processing,. He is Life Member of ISTE.
Dr. E. Sreenivasa Reddy s received B.Tech(Electronics & Communication Engineering) from
Nagarjuna University, M.S(Electronics & Control) from Birla Institute of Technology & Science,
M.Tech(Computer Engg) from Visveswaraya Technological University and Ph.D (Computer
Science & Engineering ) from AcharayaNagarjuna University ,AP, India.. He has 23 years of
teaching experience and Currently he is working as Dean, University College of Engineering,
Acharya Nagarjuna University, AP, India.. He has published more than 20 research papers in
National/ International Journals/ Conferences. His areas of research interest are Image processing,
Biometrics, Security and Micro controllers. He participate many International workshops held at
South Korea, Malaysia. He is Member of IEEE, ISTE.

An Efficient Fingerprint Identification using Neural Network and BAT Algorithm

Recommended

More Related Content

What's hot (20)

Similar to An Efficient Fingerprint Identification using Neural Network and BAT Algorithm (20)

More from IJECEIAES (20)

Recently uploaded (20)

An Efficient Fingerprint Identification using Neural Network and BAT Algorithm