2016 International Conference on Automatic Control and Dynamic Optimization Techniques (ICACDOT)

International Institute of Information Technology (I²IT), Pune

Fusion of Fingerprint, Palmprint and Iris for

Person Identification
Archana P. Patil1, D. G .Bhalke2
Department of Electronics and Telecommunication,
Rajarshi Shahu College of Engineering , Tathawade,
Savitribai Phule Pune University, Pune. Maharashtra, India.
[email protected], [email protected]

Abstract—Multibiometric system used to enhanced accuracy of Now a days, multibiometrics system is most probably
the authentication process and also reduce error rates such as used for security purpose. Multi biometrics system can be
FAR, FRR. Person identification required in many systems such developed using various techniques such as: fusion at data-
as area-access control, PC login, e-commerce etc. The biometric sensor level, matching level, feature extraction level, and
system is most probably used for security purpose. Unimodal
biometric and multimodal biometric are the two frameworks of
decision level. To evaluate the authentication decision the
biometric systems. A single biometric trait is used in unimodal maximum matching scores of all images for each person is
system, where more than one biometric traits are used in selected and then compare with threshold. If the matching
multimodal system. Multimodal framework is more precise as score is above the threshold the user is identified as genuine. If
contrast with unimodal biometric framework. In proposed the matching score is under the threshold the user is
multibiometric system three traits are combined such as authenticate and if the matching score is above the threshold
fingerprint, plamprint & iris. The standard database is used to the user is not authenticate.
evaluate the proposed system. Various features are taken from
every trait by using different features extraction algorithms. II. RELATED WORK
Matching score of these extracted features are calculated
separately. These individual scores are combines together by For high performance research is also developed on fusion of
using weighted fusion technique. As per the observation 95.23 % hyperspectral images. Recursive filtering is used on extracted
accuracy is achieved, which overcome the limitations of present features. This is done by Xudong Kang et al [1]. It reduce the
system. computational complexity and increases accuracy of
Keywords—Multimodal biometrics; weighted score level fusion; hyperspectral image classification. David Zhang et al.[2]
Minutia point extraction algorithm; Haar wavelet; Gabor filter focused on an online palm print identification, in which the
low-resolution palm print images are used. 2D gabor filter,
I. INTRODUCTION multisource image fusion method using support value
transform is proposed by Sheng Zheng et al.[3], The SVMs
Biometrics system usually works for individual person classifier is used to calculate the support values of image. 2D-
identification. In some events of unibiometric system noisy DWT is use for multi-determination highlight extraction by
data, unsatisfactory blunder rates and parody assaults such Parmeshwar Manegopale, [4]. A biometric palmprint
problems are occur. To overcome these issues multibiometric acknowledgment system used by Sumalatha K.A et al. [5], in
system is used. Combining the left and right palmprint images which palmprint images are taken using a low resolution
is also one of the popular techniques of authentication [10]. camera. The multi modal biometrics using palmprint and
Different database is used for each trait. Fingerprint images palmvein, in which wavelet based texture features extractor is
are taken from FVC database. Iris images are taken from IITD used from palm print, which is developed by V.Usharani et al.
database and palmprint images are taken from CASIA [6]. Palm print acknowledgment using ridge features are
database. 10 persons database (4 images of each person) are developed by Allen George1 et al. [7].
train for each trait. In proposed system three different
biometric traits are used, which are fingerprint, plamprint & III. PROPOSED APPROACH
iris. By using different feature extraction algorithms individual In proposed framework three biometric traits are
score is calculated for each trait. These scores are combined fused together, which are fingerprint, palm print and iris.
together using weighted fusion technique. Then this score is Different features are extracted from the palm print, iris and
used for person identification. The knn classifier is used to fingerprint. Score of extracted features are calculated by
identify the exact person. The k-NN algorithm is a statistic matching score level system and these scores are fused
classification method which store information of input dataset together using weighted sum technique. Features of test
and calculate the nearest neighbour. images and train images are compared to obtain matching

978-1-5090-2080-5/16/$31.00 ©2016 IEEE 960

scores. The last score is then used to announce the individual more information sets to extensively littler representations.
as certified or an impostor. It converts the differential equation in to a set of algebraic
equations. Haar wavelet decomposes image into K th level
where K= 1, 2, 3 etc. It calculates the horizontal, vertical,
and diagonal orientation at each level.
Haar wavelet function is defined as follows:
1
hR ( x) (4)
m

­ j/2 k 1 K 1/2
, d x d
°2 j j
2
° 2
°
° j/2 k 1/ 2 K (5)
(x) ® 2 , d x d
hi j j
° 2
2
°
0 , otherwise
°
°
¯
Where,
Fig. 1. Overview of proposed multibiometric system
m 2 j( j 0 ,1 , 2 ,... J ) indicates the level of the wavelet.
Fingerprint features are extracted by using minutiae
extractor. This calculates the separation between the edges i 0 ,1, 2 ,... m  1 is the translation parameter.
and valleys in a fingerprint image and similarly extracts
the minutiae. Minutia marking stage where mark the Gaussian filter used to calculate the center and radius of pupil.
ridged and bifurcation of fingerprint. In the edge detection Every framework gives a matching score denotes the
of fingerprint right and left both edges are detected. Left proximity of the feature vector with the template vector. These
edge detected using, values are combined together using weighted fusion
technique. These fused score then compare with threshold
 x my  b   value. The kNN classifier used to know the person identity.k
NN is a case-based learning method. All 40 images database is
Right edge detected using,
including in training dataset. It measures the Euclidean
y mx  b (2) distance between each point and find the nearest point.

where, m represents slop.

d 2 st ( x s  y )( x s  y ) '
(6)
Plamprint elements are extracted using gabor filter. t t
2D Gabor filter is used to calculate the texture value of
palmprint image. Texture features are calculated at Where various distances between xs and y are measures.
t
different frequencies and orientations of palmprint.
The value of K is automatically determined depending on
(
'2
 y
2
. y
2
)
different data.
G (x, y) exp
x
cos( 2 3
x'
)
(3)
2
 2 y
V
IV. RESULT
where,
x'=x cosϴ+y sinϴ, y' = - x.sinϴ +y cosϴ A. Minutia points extraction from fingerprint
σ is the variance
λ is the wavelength
γ is the aspect ratio
θ is the orientation.

Iris features are taken using haar wavelet transform, which

is one of the easiest wavelet transforms which can convey

961
Fig. 2. Minutia points extraction from fingerprint

Minutia points are taken from the input fingerprint image.

Minutiae points are the nearby edge discontinuities, which are
of two sorts: ridge endings and bifurcations.
B. Texure value extraction from palmprint

Fig. 3. Texture value evaluation from palmprint

Gabor filter used to extract the palmprint features. It calculates Fig. 5. Iris feature extraction
the texture value of image. Texture value measures at five
frequencies and eight orientations of extracted features. Edges Haar wavelet is used for iris feature extraction. It transforms
are also calculated at different local discontinuity. huge data sets in to representations. It de-composes the images
in different level using wavelet transform. The energy measure
is given as,
C. Different orientation extraction of iris
y
 E ¦ si ( j , k )^ 2  
j,k

Here wavelet energies of each iris sub images are used to

calculate threshold to encode the sub images. The threshold
[T] is given as,
P ( E 1 ........ En )
 K  
Max ( E 1 .......... . En )

Where K is constant, E1… En are appropriate wavelet sub

image energies of 1….n sub images and μ (E1……..En) are
mean wavelet peak energy values. After extraction of features
from all three traits, these scores are fused and classify using
knn classifier.The k-NN classifier is the simplest classification
Fig. 4. Edge detection of iris method where asymptotically tends to the optimal Bayes
classifier under mild hypotheses on k and N. Since a proper
Iris of the eye is the colored area that surrounds pupil. After training phase is not required. A euclidean distance is
taking input iris image delete the extra portion and then calculated between two nearest vectors. Accuracy of whole
histogram is calculated. The Gaussian filter clears the noise of system is calculated by observation of false acceptance rate
image using the 2D distribution function. After that canny and false reject rate of total system. Scores are used to
edge detector is used to detect the edges of images. Fig. 4 represent the higher similarity between the templates. Then
shows the vertical, horizontal and diagonal representation at single threshold is set to differ between client and imposer.
first level.
D. FAR and FRR Calculation
Below observations shows the FAR and FRR of all three
biometric traits separately and Fig. d. shows the FAR and FRR
of proposed fused system.

962
 V. CONCLUSION
The multibiometric system used to overcome the
problems of unibiometrics. Here the accuracy of unimodal
biometric system using fingerprint, palmprint and iris,
obtained is 72.73%, 65.57%, and 80% respectively. The
accuracy of the proposed system obtained is 95.23%.
Proposed system develops a fusion at matching score
level, which is a fastest fusion. The proposed system
demonstrates the scores of three traits. These scores measure
the similarity among the traits. Scores are combined together
using weighted fusion technique. By experimental results it
can be concluded that multimodal system is precise than
(a) (b) unimodal system.

REFERENCES

[1] Xudong Kang, Shutao Li and Jón Atli Benediktsson, “Feature Extraction
of Hyperspectral Images with Image Fusion and Recursive Filtering”,
IEEE trans. on geosciences and remote sensing, vol. 52, no. 6, June
2014.
[2] David Zhang, Wai-Kin Kong, Jane You, and Michael Wong, “Online
Palmprint Identification”, IEEE trans. on pattern analysis and machine
intelligence, vol. 25, no. 9, Sept. 2003.
[3] Sheng Zheng, Wen-Zhong Shi, Jian Liu, Guang-Xi Zhu, and Jin-Wen
Tian, “Multisource Image Fusion Method Using Support Value
Transform”, IEEE trans. on image processing, vol. 16, no. 7, July 2007.
[4] Parmeshwar Manegopale, “A Survey on Palm print Recognition”,
International Journal of Innovative Research in Science, Engineering
(c) (d) and Technology, vol. 3, issue 2, February. 2014.
Fig. 6. (a) FAR and FRR of Fingerprint (b) FAR and FRR of Palmprint (c) [5] Sumalatha K.A, Harsha H, “Biometric Palmprint Recognition System”,
FAR and FRR of Iris (d) Accuracy FAR and FRR A Review International Journal of Advanced Research in Computer
Science and Software Engineering, vol. 4, issue 1, January 2014.
[6] V.Usharani, S.V.Saravanan, “Multi Modal Biometrics Using Palmprint
From below results it is shown that accuracy of fused system and palmvein”, Journal of Theoretical and Applied Information
is 95.23% , which is greater than each individual trait. Technology, 10th vol. 67 no.1, Sept. 2014. .
[7] Allen George1, G.Karthick, “Palmprint Recognition Using Ridge
Features”, International Journal of Advanced Research in Computer and
TABLE I. ACCURACY CALCULATION Communication Engineering, vol. 2, issue 11, Nov. 2013.
[8] Yong Xu, Lunke Fei, and David Zhang, "Combining Left and Right
Palmprint Images for More Accurate Personal Identification", IEEE
Sr No Biometric Database Accuracy trans. on image processing, vol. 24, no. 2, Feb. 2015.
trait [9] Karthikeyan.T, Sumathi.T.K, "Implementation of Biometric Personal
1 Fingerprint FVC 72.73% Identification based on Normalized Approach of Fusion technique",
2 Plamprint IITD 65.57% International Journal of Advanced Information in Arts Science &
3 Iris CASIA 80.00% Management (IJAIASM) ISSN, vol.4, no.8, April 2015.
4 Fused system FVC+ IITD+ 95.23% [10] Roli Bansal, Priti Sehgal and Punam Bedi, "Minutiae Extraction from
CASIA Fingerprint Images - a Review", IJCSI International Journal of
Computer Science Issues, vol. 8, issue 5, no 3, Sept. 2011.
[11] Paithane, A. N., D. S. Bormane, and Ujwala Patil. "Novel Algorithm for
Feature Extraction and Feature Selection from Electrocardiogram
As mentioned above, these results show a marked Signal." International Journal of Computer Applications 134.9 (2016): 6-
improvement in accuracy over the individual unimodal 9.
systems. Proposed implemented system is highly accurate than [12] Xuanbin Si, Jianjiang Feng, Jie Zhou, and Yuxuan Luo, "Detection and
Rectification of Distorted Fingerprints"IEEE trans. on pattern analysis
unimodal system. Different databases are used for each trait as and machine intelligence, vol. 37, no. 3, March 2015
shown above. 10 persons images (4 images of each person) of
each traits are taken from respective database.

963