2.human verification using multiple fingerprint texture 7 20

Computer Engineering and Intelligent Systems www.iiste.org
ISSN 2222-1719 (Paper) ISSN 2222-2863 (Online)
Vol 2, No.8, 2011

Human Verification using Multiple Fingerprint Texture
Matchers

Zahoor Ahmad Jhat
1
Department of Electronics, Islamia College of Science and commerce, Srinagar, J&K (India)
zahoorjhat@gmail.com

Ajaz Hussain Mir
Department of Electronics & Communication Engineering, National Institute of Technology, Hazratbal,
Srinagar, J&K (India)
ahmir@rediffmail.com

Seemin Rubab
Department of Physics, National Institute of Technology, Hazratbal, Srinagar, J&K (India)
ask_rubab@yahoo.co.in

Received: 2011-10-23
Accepted: 2011-10-29
Published:2011-11-04

Abstract
This paper presents a multimodal biometric verification system using multiple fingerprint matchers. The
proposed verification system is based on multiple fingerprint matchers using Spatial Grey Level
Dependence Method and Filterbank-based technique. The method independently extract fingerprint
texture features to generate matching scores. These individual normalized scores are combined into a
final score by the sum rule and the final score is eventually used to effect verification of a person as
genuine or an imposter. The matching scores are used in two ways: in first case equal weights are assigned
to each matching scores and in second case user specific weights are used. The proposed verification
system has been tested on fingerprint database of FVC2002. The experimental results demonstrate that the
proposed fusion strategy improves the overall accuracy of the system by reducing the total error rate of the
system.
Keywords: - Multimodal biometric System, Fingerprint verification, SGLDM, Filterbank matching, Score
level fusion, Sum rule.

1. Introduction
In today’s wired information society when our everyday life is getting more and more computerized,
automated security systems are getting more and more importance. The key task for an automated security

7|Page
www.iiste.org

Vol 2, No.8, 2011
system is to verify that the users are in fact who they claim to be. Traditionally password and ID cards have
been used for human verification to restrict access to secure systems such as ATMs, computers and security
installations [1]. The drawback with the traditional systems is that a password can be guessed or forgotten
and similarly the ID card can be lost or stolen, thus rendering such methods of human verification
unreliable. To overcome these problems, biometrics offers an alternative. Biometrics refers to identifying a
person based on his or her physiological or behavioral traits. Face, fingerprints, hand geometry, iris, retina,
signature, voice, facial thermogram, hand vein, gait, ear, odor, keystroke, etc. are some of the biometric
features that are used for human verification and identification. Most of the biometric systems that are in
use in practical application use a single piece of information for recognition and are as such called
unimodal biometric systems. The unimodal biometric recognition systems, however, have to contend with
a variety of problems like non-universality, susceptibility to spoofing, noise in sensed data, intra-class
variations, inter-class similarities. Some limitations of the unimodal biometric systems can be alleviated
by using multimodal system [2]. A biometric system that combines more than one sources of information
for establishing human identity is called a multimodal biometric system. Combining the information cues
from different biometric sources using an effective fusion scheme can significantly improve accuracy [3] of
a biometric system.
The information fusion in multibiometrics can be done in different ways: fusion at the sensor level,
feature extraction level, matching score level and decision level. Sensor level fusion is rarely used as fusion
at this level requires that the data obtained from the different biometric sensors must be compatible, which
is seldom the case. Fusion at the feature extraction level is not always possible as the feature sets used by
different biometric modalities may either be inaccessible or incompatible. Fusion at the decision level is too
rigid as only a limited amount of information is available. Fusion at the matching score level is, therefore,
preferred due to presence of sufficient information content and the ease in accessing and combining match
scores [4].

2. Related work
A number of works showing advantages of multimodal biometric verification systems have been reported
in literature. Brunelli and Falavigna [2] have proposed personal identification system based on acoustic and
visual features, where they use a HyperBF network as the best performing fusion module. Duc et al. [5]
proposed a simple averaging technique combining face and speech information. Kittler et al. [6] have
experimented with several fusion techniques using face and voice biometrics, including sum, product,
minimum, median, and maximum rules and they have found that the best combination results are obtained
for a simple sum rule. Hong and Jain [7] proposed a multimodal personal identification system which
integrates face and fingerprints that complement each other. The fusion algorithm combines the scores from
the different experts under statistically independence hypothesis. Ben-Yacoub et al. [8] proposed several
fusion approaches, such as Support Vector Machines (SVM), tree classifiers and multi-layer perceptrons,
combining face and voice biometrics. Pigeon at el. [9] proposed a multimodal person authentication
approach based on simple fusion algorithms to combine the results coming from face, and voice biometrics.
Choudhury et al. [10] proposed a multimodal person recognition using unconstrained audio and video and
the combination of the two features is performed using a Bayes net. Jain at el. [11] combine face,

8|Page
www.iiste.org

Vol 2, No.8, 2011
fingerprint and hand geometry biometrics combining them under sum, decision tree and linear
discriminant- based method. The sum rule is reperted to outperform others. Various other biometric
combinations have been proposed [12, 13, 14] that report that combining more than one biometric
modalities together result in improved performance than using them alone. Jhat et al. [15 ] have proposed
a unimodal fingerprint biometrics verification system using texture feature of Energy of a
fingerprint as a biometric trait that gives 70% Genuine Accept Rate (GAR) at 1% False Accept Rate
(FAR) for effecting personal verification. To augment performance of the said proposed unimodal
fingerprint verification system using a single matching score, in the present work, a multimodal biometric
system based on multiple fingerprint matchers is proposed. The use of the proposed combination strategy in
combining multiple matchers significantly improves the overall accuracy of the fingerprint based
verification system by reducing the total error rates. We have chosen multiple fingerprint matchers as
they form a good combination for a multimodal biometric system because the fusion of this combination in
such systems demonstrates substantial improvement in recognition [3, 6]. It is due to the fact that the
sources are fairly independent [16]. They not only address the problem of non-universality, since multiple
traits ensure sufficient population coverage but also deter spoofing since it would be difficult for an
imposter to spoof multiple biometric traits of a genuine user simultaneously. A multimodal biometric
verification system based on multiple fingerprint matchers is, therefore, described in this paper. To
construct the multimodal biometric verification system, we have combined two fingerprint matchers
of Spatial Grey Level Dependence Method (SGLDM) [27] and Filterbank-based [19] for extracting
matching scores. Such a system has, hitherto, not been tried in the reported literature. The rest of the
paper is arranged as follows: Section 3 describes Fingerprint verification modules. Section 4 presents
normalization of matching scores. Fusion of the normalized scores is addressed in section 5. Experimental
results are shown in section 6 and section 7 concludes the paper.

3. Verification Modules
Fingerprint is the pattern of ridges and valleys on the tip of a finger and is used for personal
verification of people. Fingerprint based recognition method because of its relatively outstanding
features of universality, permanence, uniqueness, accuracy and low cost has made it most popular
and reliable technique. Current fingerprint recognition techniques can be broadly classified as
Minutiae-based, ridge feature-based, correlation-based [17] and gradient based [18]. The
minutiae-based methods are widely used in fingerprint verification but do not utilize a significant
component of the rich discriminatory information available in the ridge structures of the fingerprints.
Further, minutiae-based methods have to contend with the problem of efficiently matching two
fingerprint images containing different numbers of unregistered minutiae points. This is the due to
these reasons that present work uses Texture- based representation of a fingerprint as the smooth flow
pattern of ridges and valleys in a fingerprint can be also viewed as an oriented texture pattern [17].

Texture has been successfully used in extracting hidden information in medical images such as
ultrasound [20], MRI [21], CT [22], retina [23] and Iris [24]. Although there is no strict definition of

9|Page
www.iiste.org

Vol 2, No.8, 2011
the image texture, however, being defined as a function of the spatial variation in pixel intensities
(grey values), is useful in a variety of applications, e.g, recognition of image regions using texture
properties [25]. Texture methods can be broadly categorized as: statistical, structural, modal, transform
[25, 26]. Teceryan et al [25] and Matreka et al [26] present review of these methods.The two texture
based matchers of Spatial Grey Level Dependence Method (SGLDM) and Filterbank-based, that are
used in the present work for personal verification, are summarized as follows:

3.1 SGLDM- based Matching
Jhat et al. [15] have used Harlicks spatial grey level dependence matrix (SGLDM) [27] method for
extracting statistical texture features. In SGLDM, second order joint conditional probability density

function, f (i, j d , θ ) for directions θ = 0, 45, 90, 135, 180, 225, 270, and 315 degrees is estimated.

Each f (i, j d , θ ) is the probability of going from grey level i to grey level j, given that the

inter-sample spacing is d and the direction is given by the angle θ. The estimated value for these
probability density functions can thus be written in the matrix form:

φ (d , θ ) = [ f (i. j d , θ )] (1)

Scanning of the image in four directions viz; θ = 0, 45, 90, 135 degrees is sufficient for computing
these probability distribution function, as the probability density matrix for the rest of the directions
can be computed from these four basic directions. This yields a square matrix of dimension equal to
the number of intensity levels in the image for each distance d and direction θ. Due to the intensive
nature of computations involved, often only the distances d= 1 and 2 pixels with angles θ = 0, 45, 90,
135 degrees are considered as suggested [26].

Let φ ′(d , θ ) denote transpose of the matrix φ (d , θ ) for the intersampling spacing, d, and direction θ.

φ (d ,0) = φ ′(d ,180)
φ (d ,45) = φ ′(d ,225) (2)
φ (d ,90) = φ ′(d ,270)
φ (d ,135) = φ ′(d ,315)

The knowledge of φ (d ,180), φ (d ,225), φ (d ,270), φ (d ,315) , add nothing to the characterization of

texture. If one chooses to ignore the distinction between opposite directions, then symmetric probability
matrices can be employed and then the spatial grey level dependence matrices

S o (d ), S 45 (d ), S 9 o (d ), S135 (d ) , can be found from

10 | P a g e
www.iiste.org

Vol 2, No.8, 2011

S o (d ) =
1
[φ (d ,0) + φ (d ,180)] = 1 [φ (d ,0) + φ ′(d ,0)] (3)
2 2

S 45 (d ) = [φ (d ,45) + φ (d ,225)] = [φ (d ,45) + φ ′(d ,45)]
1 1
Similarly
2 2
(4)

S 9 o (d ) and S135 (d ) can be similarly calculated.
Approximately two dozen co-occurrence features can be obtained using the above method and the
consideration of the number of distance angle relations also will lead to a potentially large number of
dependent features. Jhat et al. [15] have shown that the fingerprint texture feature of Energy can provide
useful information for pattern recognition and can be used for verification. The Energy texture feature
of a fingerprint is given by Equation 5.

NG −1NG −1 2

E (Sθ (d )) = ∑ ∑ [Sθ (i, j ) d ]
i =0 j =0
(5)

Where Sθ i, j d ( ) is the (i, j) th element of Sθ (d ) and N G is the number of grey levels in the image

from which the spatial grey level dependence matrices are extracted.
The texture feature of Energy of the fingerprint is calculated using algorithm of SGLDM by taking
d=1 [26], for different values of θ, for a fingerprint being a soft texture [17] require small values of
d. The results of Energy values for the angle of 0, 45, 90 and 135 degrees are obtained as shown in
Figure 1 and are used for discrimination of individuals and effecting personal verification. If the
Euclidean distance between two Energy values of query and gallery fingerprint image is less than a
threshold, then the decision that the two images belong to same finger is made, alternately a decision
that they belong to different fingers is made.

Energy values at angle 0
9010000
8010000
Energy Values

7010000
6010000
5010000
4010000
3010000
2010000
1010000
10000
0 20 40 60 80 100

Fingerprints

11 | P a g e
www.iiste.org

Vol 2, No.8, 2011
9010000
Energy Values

8010000
7010000
6010000
5010000
4010000
3010000
2010000
1010000
10000
0 20 40 60 80 100

Fingerprints

9010000
Energy Values

8010000
7010000
6010000
5010000
4010000
3010000
2010000
1010000
10000
0 20 40 60 80 100

Fingerprints


9010000
8010000
Energy Values

7010000
6010000
5010000
4010000
3010000
2010000
1010000
10000
0 20 40 60 80 100

Fingerprints

Figure 1. Energy Values for the angles of 0, 45, 90 and 135 degrees.

3.2 Filterbank-based Matching
Jain et al. [19] have proposed a fingerprint representation scheme that utilizes both global and local
features in a compact fixed length feature vector called ‘FingerCode’. The proposed scheme makes use of
the texture features available in a fingerprint to compute the feature vector. In the Filter-based matching,
generic representation of oriented texture relies on extracting a core point in the fingerprint which is
defined as the point of maximum curvature of the ridge in a fingerprint. Then a circular region around the
core point is located and tessellated into sectors. The pixel intensities in each sector are normalized to a

12 | P a g e
www.iiste.org

Vol 2, No.8, 2011
constant mean and variance and filtered using a bank of 8 Gabor filters to produce a set of 8 filtered images.
Grayscale variance within a sector quantifies the underlying ridge structures and is used as a feature. A
feature vector termed as a FingerCode, is the collection of all the features, computed from all the sectors, in
every filtered image. The FingerCode captures the local information, and the ordered enumeration of the
tessellation captures the invariant global relationships among the local patterns. The matching stage simply
computes the Euclidean distance between the two corresponding FingerCode values. Figure 2 depicts
diagrammatic representation of the Filterbank matching algorithm as proposed by Jain et al. [19].

Figure 2. Diagrammatic representation of Filterbank Matching algorithm.

The first two steps of determining a center point for the fingerprint image and tessellate the
region around the center point are straightforward. The filtering process and obtaining of feature vector can
be summarized as follows:

Let I x, y denote the gray value at pixel x, y in an M N fingerprint image and let M and V ,
3.2.1 Filtering

the estimated mean and variance of sector S , respectively, and N x, y , the normalized gray-level value
at pixel x, y . For all the pixels in sectorS , the normalized image is defined as:

M , if I x, y M
,

N x, y
M ! , otherwise
,
(6)

Where M) and V) are the desired mean and variance values, respectively. The values of both M) and
V) have been set to 100.
An even symmetric Gabor filter has the following general form in the spatial domain:

G x, y; f, Ѳ exp . 1 67 cos 2πfx ,
/ 2 2

0 342 352
(7)

13 | P a g e
www.iiste.org


x ; = x sin Ѳ y cos Ѳ
Vol 2, No.8, 2011

y ; = x cos Ѳ ! y sin Ѳ
(8)

Where f is the frequency of the sinusoidal plane wave along the direction Ѳ from the x-axis, and δ 2 and
(9)

δ 2 are the space constants of the Gaussian envelope along x, and y, axes, respectively. Let H indicate the
enhanced image. Convolving H with eight Gabor filters in the spatial domain would be a computationally

F(H) denote the discrete Fourier transform of H, and F GѲ indicate the discrete Fourier transform of the
intensive operation. To speed up this operation the convolution is performed in the frequency domain. Let

Gabor filter having the spatial orientation Ѳ. Then the Gabor filter image,VѲ , may be obtained as,
/ ?F
VѲ = F H F GѲ A
Where F
(10)
/
is the inverse Fourier transform. Eight filtered images are obtained in this way.

3.2.2 Feature vector

Let C Ѳ x, y be the component image corresponding to Ѳ for sector S . For ∀ i ,i=0,1………………,47
The standard deviation within the sectors, in the filter-bank algorithm,, define the feature vector.

(as total of 48 sectors from S) to SCD are defined in six concentric bands around the central point) and
Ѳ ϵ [ 00, 450, 900, 1350] ( as a fingerprint image is decomposed into four components images corresponding
to four different values of Ѳ as mentioned). A feature is standard deviation FiѲ , which is defined as :

F Ѳ = ∑ (C θ ( x , y ) − M θ ) 2
i i
ki
(11)
Where Ki is the number of pixels in Si and MiѲ is the mean of the pixel values in CiѲ(x,y) in sector S . The
average absolute deviations of each sector in each of the eight filtered image define the components of the
feature vector called FingerCode. Fingerprint matching is then based on finding the Euclidean distance
between the corresponding FingerCodes.

4. Normalization.
Normalization involves transforming the raw scores of different modalities to a common domain using
a mapping function. In our case, both the matching scores are distance scores, yet they have different
numerical range. To transform these numerically incompatible matching scores into a common domain
prior to fusion, normalization is needed. Comparing different normalization techniques on different
multimodal biometric systems, Ribaric at el. [28] conclude that no single normalization technique performs
best for all systems. We have, therefore, used min-max technique. This technique is not only simple but
best suited for the case where maximum and minimum values of the scores produced by the matcher are
known. Besides, minimum and maximum scores can be easily shifted to 0 and 1 respectively. The
matching scores are normalized using min-max technique as follows.

Let G represent the gallery templates, Q represent the query samples, and S qg represent the

match score of the particular query ‘q’,qϵQ with gallery template ‘g’, gϵG. Then S qG represents the
vector of scores obtained when a query ‘q’ is matched against the entire gallery G. In min-max

14 | P a g e
www.iiste.org

Vol 2, No.8, 2011
normalization, the minimum and the maximum of this score vector are used to obtain the normalization

score S' qg as per Equation12. The normalized score lie in the range 0-1.

S qg − min(S qG )
S ' qg =
max(S qG ) − min(S P G )
(12)

5. Fusion.
The matching scores, next to feature vectors, output by matchers contain the richest information [4]
about the input pattern. Further, it is relatively easy to access and combine the scores generated by the
different matchers. Consequently, integration of information at the matching score level is the most
common approach in the multimodal biometric systems. The proposed method, therefore, fuses the
individual match scores of the fingerprint and the fused score is used for verification. There are several
classifiers for the fusion and analysis of several classifier rules is given in [6, 11]. It is suggested that the
weighted sum rule is more effective and outperforms other fusion strategies based on empirical
N
observations [29]. The weighted sum rule is defined as S fusion = ∑ Wi S i Where S i is the normalized
i =1

matching score provided by the i th trait and W i is the weight assigned to the i th trait. The identity of a

person is verified if S fusion ≥ η , where η is the matching threshold. The weighting of the matching
scores has been done in the following ways:

5.1 Weighing Matching Scores Equally

In the first experiment, equal weightage is given to two matching scores of a fingerprint and a new
2
1
score is obtained. Then the final matching score S fusion = ∑ S i is compared against a certain
i =1 2

threshold value to make a decision for a person being genuine or an imposter. The Figure 3a shows the
improved matching performance when equal weightage is given to both matching scores of the
fingerprint.

5.2 Weighing Matching Scores Unequally
When biometric trait of a user cannot be reliably acquired, the user will experience high false reject
rate. This can result when the biometric trait becomes unreadable due to dirty or worn down dry fingers. In
such a situation, the false error rate can be reduced and accuracy improved if different matching scores are
weighted differently for increasing the influence of one or the other matching score as per degree of
importance for different users. Weights indicate the importance of individual biometric matchers in a
multibiometric system and, therefore, the set of weights are determined for a specific user such that the

15 | P a g e
www.iiste.org

Vol 2, No.8, 2011
total error rates corresponding to that user can be minimized. User specific weights are estimated [29] from
the training data as follows:

1. For the i th user in the database, vary weights W1,i , and W2 ,i over the range [ 0,1], with

the condition W1,i + W2,i =1

2. Compute S fusion = W1,i S1 + W2,i S 2
3. Choose that set of weights that minimizes the total error rate associated with the scores. The total
error rate is sum of the false accept and false reject rates.
The user specific weight procedure utilizes the histograms of both the genuine and imposter score and
computing user-specific thresholds using imposter scores have been shown not to improve performance [30]
very much. In the second experiment, with a common threshold, therefore, we assign different weights
to matching scores to minimize false accept rate and false reject rate associated with an individual and
improve further the matching performance. The improved matching performance when user specific
weights are used, is shown in Figure 3b.

6. Experimental Results
The suggested method has been tested on fingerprint databases of FVC2002 DB1 and DB2 [31]. Both
the databases contain images of 110 different fingers with 8 impressions for each finger yielding a total of
880 fingerprints in each database. The databases has been divided into two sets: A and B. Set A
contains the fingerprint images from the first 100 fingers as evaluation set and Set B contains the
remaining 10 fingers as a training set. About 10 fingerprint images were eliminated from the database as
Filter-based matcher rejected the images either being of poor quality or failing to locate the center. The
False Accept Rate (FAR) and False Reject Rate (FRR) for the suggested method were evaluated by using
the protocols of FVC2002 [32]. Each fingerprint impression in the subset A is matched against the
remaining impressions of the same finger to compute genuine distribution. The total genuine attempts is
(8×7)/2×90 = 2520. For Imposter distribution, the first fingerprint impression of each finger in subset A is
matched against the first impression of the remaining fingers. The total imposter attempts is (90×89)/2 =
4005. The normalized genuine and imposter distribution matching scores for DB1 and DB2 are shown in
Figures 4(a) and (b) respectively.
For the multiple matcher combination, we randomly selected each of the genuine and imposter
scores for the training and remaining each half for the test. This process has been repeated 5 times to give 5
different training sets and 5 corresponding independent test sets. For authentication, we randomly selected
four impressions of each fingerprint and enrolled them as templates into the system database. The
remaining 90 × 4 = 360 fingerprints images in each database were used as input fingerprints to test the
performance of our proposed method. The matching scores of the two classifiers are then summed and the
final matching score is compared against a certain threshold value to recognize the person as genuine or an
imposter. The FAR and FFR rates with different threshold values were obtained based on 90 × 360 = 32400
matches in each database.

16 | P a g e
www.iiste.org

Vol 2, No.8, 2011
False Accept Rate False Reject Rate
FAR (%) FRR (%)
SGLDM Filter SGLDM +Filter
1 19.8 15.3 4.9
.1 34.5 26.0 13.8
.01 39.4 32.1 17.3
Table 1. False Reject Rates ( FRR) with different values of False accept rates (FAR) when matching
scores are equally weighted.
False Accept Rate False Reject Rate
FAR (%) FRR (%)
SGLDM Filter SGLDM +Filter
1 18.2 14.5 3.8
.1 33.2 24.9 12.7
.01 37.8 30.9 15.5
Table 2. False Reject Rates ( FRR) with different values of False accept rates (FAR) when matching scores
are unequally weighted.
To demonstrate the effectiveness of the proposed method, tables showing FAR and FRR are drawn in Table
1 and Table 2. Besides, ROC curves between FAR and GAR have also been plotted in Figure 3a and Figure
3b. It is evident from the ROC curves that performance gain obtained for the proposed fusion system
is higher as compared to the performance obtained for two individual matchers. As shown in the
Figures 3a and 3b, the integration of matchers enhances the performance of the proposed multimodal
verification system over the unimodal fingerprint matcher as proposed in [15] by giving Genuine Accept
Rate ( GAR ) of 95.1% and 96.2% respectively at False Accept Rate ( FAR ) of 1%.

7. Conclusion
A biometric verification system using a single fingerprint texture matcher is less accurate for
effecting personal verification. To enhance the performance of such a unimodal verification system, a
multimodal biometric verification system using multiple fingerprint matchers is proposed. The proposed
verification system use Spatial Grey Level Dependence Method (SGLDM) and Filterbank-based matching
technique to independently extract fingerprint texture features to generate matching scores. These
individual normalized scores are combined into a final score by the sum rule. The matching scores are used
in two ways, in first case equal weights are assigned to each matching scores and in second case user
specific weights are used. The final fused score is eventually used to conclude a person as genuine or an
imposter. The proposed verification system has been tested on fingerprint database of FVC2002. The
experimental results demonstrate that the proposed fusion strategy improves the overall accuracy of the of
the unimodal biometric verification system by reducing the total error rate of the system.

17 | P a g e
www.iiste.org

Vol 2, No.8, 2011

SGLDM
Genuine Accept Rate (%)

Filter
SGLDM + Filter

False Accept Rate (%)

Figure 3. ROC curves showing performance improvement of combination of matchers over individual
matchers when matching scores are (a) weighted equally (b) weighted unequally.

Genuine Imposter
Genuine
Percentage
Percentage

Threshold Threshold

(a) (b)
Figure 4. Genuine and Imposter distributions for (a) DB1, (b) DB2.

References
[1]. M. S. Khalil, M. DZulkifli, M. K. Khan and Q.AL-Nuzaili (2010), Fingerprint verification using
statistical descriptors, Digital Signal Processing, 20, pp.1264-1273.

18 | P a g e
www.iiste.org

Vol 2, No.8, 2011
[2]. R. Brunelli, and D. Falavigna (1995), Personal Identification Using Multiple Cues, IEEE Trans.
on Pattern Analysis and Machine Intelligence, 17, pp. 955-966.
[3]. L. Hong, A. Jain and S. Pankanti (1999), Can Multibiometrics Improve Performance?, in
Proceedings AutoID'99, Summit, NJ, pp. 59-64.
[4]. A. Ross (2007), An Introduction to Multibiometrics. in Proceedings of the 15th European Signal
Processing conference (EUSIPCO),(Pozan, Poland).
[5]. B. Duc, G. Maýtre, S. Fischer and J. Bigun (1997), Person Authentication by Fusing Face and
Speech Information, in Proceedings of the First International Conference on Audio and
Video-based Biometric Person Authentication, March, 12-24, Crans-Montana, Switzerland,
311-318.
[6]. J. Kittler, M. Hatef, R. P. W. Duin, and J. Matas (1998), On Combining Classifiers, in IEEE
Transactions on Pattern Analysis and Machine Intelligence, 20, 226–239.
[7]. L. Hong and A. Jain (1998), Integrating Faces and Fingerprints for Personal Identification”, in
IEEE Transactions on Pattern Analysis and Machine Intelligence, 20, 1295–1307.
[8]. S. Ben-Yacoub, Y. Abdeljaoued, and E. Mayoraz (1999), Fusion of Face andSpeech Data for
Person Identity Verification, IEEE Trans. Neural Networks, 10, 1065-1075.
[9]. S. Pigeon and L. Vandendorpe (1998), Multiple Experts for Robust Face Authentication”, in SPIE,
Editor, Optical Security and Counterfeit Deterrence II 3314, 166-177.
[10]. T. Choudhury, B. Clarkson, T. Jebara and A. Pentland, (1999), Multimodal Person Recognition
Using Unconstrained Audio and Video”, in Second International Conference on Audio- and
Video-based Biometric Person Authentication, March, 22-23, Washington D.C., USA, 176–181.
[11]. A. Ross and A.K. Jain (2003), Information Fusion in Biometrics, Pattern Recognition Letters, 24,
2115-2125.
[12]. Anil K. Jain, Karthik Nandakumar, Xiaoguang Lu, and Unsang Park (2004), Integrating Faces,
Fingerprints, and Soft Biometric Traits for User Recognition, Proceedings of Biometric
Authentication Workshop, LNCS 3087, Prague, 259-269.
[13]. C. Chen, C. Chu (2005), Fusion of Face and Iris Features for Multimodal Biomertics, Springer
Verlag, Berlin Heidelberg.
[14]. M. Nageshkumar, P. K. Mahesh and M. N. Shanmukha Swamy (2009), An Efficient Secure
Multimodal Biometric Fusion using Palmprint and Face Images, IJCSI International Journal of
Computer Science Issues, 2,
[15]. Z. A. Jhat; A. H. Mir; S. Rubab (2011), Personal Verification using Fingerprint Texture Feature '
International Journal of Security and its Applications, 5.
[16]. M. Indovina, U. Uludag, R. Snelick, A. Mink, and A. Jain (2003), MultimodalBiometric
Authentication Methods: A COTS Approach,” Proc. MMUA 2003,Workshop Multimodal User
Authentication, 99-106.
[17]. S. Chikkerur, S. Pankanti, A. Jea and R. Bolle (2006), Fingerprint Representation using Localized
Texture Features, The 18th International Conference on Pattern Recognition.

19 | P a g e
www.iiste.org

Vol 2, No.8, 2011
[18]. G. Aggarwal, N. K. Ratha, Tsai-Yang Jea and R. M. Bolle (2008), Gradient based textural
characterization of fingerprints”, in proceedings of IEEE International conference on Biometrics:
Theory, Applications and Systems.
[19]. A. K. Jain, S. Prabhkar (2000), Hong L., Pankanti, S.: Filterbank-Based Fingerprint Matching, IEEE
Transactions on Image Processing, 9, 846-853.
[20]. M. F. Insana, R. F. Wanger, B. S. Garra, D. G. Brown, T. H. Shawker(1986), Analysis of Ultrasound
Image Texture Via Generalized Rician Statistics, Optical Engineering, 25, 743-748.
[21]. R. Lerski, K. Straughan, L. Shad, D. Boyce, S. Bluml and I. Zuna (1993), MR Image Texture
Analysis-An Approach to Tissue Characterization, Magnetic Resonance Imaging, 873-887.
[22]. A. H. Mir, M. Hanmandlu and S. N. Tandon(1995), Texture Analysis of CT Images, IEEE
Engineering in Medicine and Biology, 781-786.
[23]. K. Yogesan, R. H. Eikelboom, C. J. Barry (1998), Texture Analysis of Retinal Images to
Determine Nerve Fibre Loss, Proceedings of Fourteenth International Conference on Pattern
Recognition, 2, 1665-1667.
[24]. L. Ma, T. Tan, Y. Wang and D. Zhang (2003), Personal Identification Based on Iris Texture
Analysis. IEEE Computer Society, 25, 1519-1533.
[25]. M. Tuceryan, A. K. Jain (1998), Texture Analysis, The Handbook of Pattern Recognition and
Computer Vision (2nd Edition) by Chen, C. H., Pau, L. F., Wang, P. S. P., World Publishing Co,
207-248.
[26]. A. Matrerka and M. Strazelecki (1998), Texture Analysis Methods-A Review. Technical University
of Lodz, Institute of Electronics, COST B11 report, Brussels.
[27]. R. M. Haralick (1979), Statistical and Structural Approaches to Texture, Proceeding of the IEEE,
67, 786-804.
[28]. S. Ribaric and I. Fratric, (2006), Experimental evaluation of matching-scorenormalization
techniques on different multimodal biometric systems.In: IEEE Mediterranean Electrotechnical
Conf., 498–501.
[29]. A.K. Jain and A. Ross (2004), Multibiometric systems, Communications of the ACM,Special Issue
on Multimodal Interfaces, 47, 34-40.
[30]. A.K. Jain and A. Ross (2002), Learning User-Specific Parameters in a MultibiometricSystem, Proc.
IEEE Int’l Conf. Image Processing, 57-60.
[31]. FVC2002 https://bias.csr.unibo.it/fvc2002
[32]. R. Cappelli, D. Maio, D. Maltoni, J. L. Wayman, A. K. Jain (2006), Performance Evaluation of
Fingerprint Verification Systems, IEEE Transactions on Pattern Analysis and Machine Intelligence,
28.

20 | P a g e
www.iiste.org

2.human verification using multiple fingerprint texture 7 20

More Related Content

What's hot (19)

Similar to 2.human verification using multiple fingerprint texture 7 20 (20)

More from Alexander Decker (20)

Recently uploaded (20)

2.human verification using multiple fingerprint texture 7 20