Welcome to Scribd, the world's digital library. Read, publish, and share books and documents. See more
Download
Standard view
Full view
of .
Save to My Library
Look up keyword
Like this
1Activity
0 of .
Results for:
No results containing your search query
P. 1
PIFS Code Base for Biometric Palmprint Verification

PIFS Code Base for Biometric Palmprint Verification

Ratings: (0)|Views: 132 |Likes:
Published by ijcsis
This paper proposes a new technique to extract the palmprint features based on some fractal codes. The palmprint features representation is formed based on position of range blocks and direction between the position of range and domain blocks of fractal codes. Each palmprint representation is divided into a set n blocks and the mean value of each block are used to form the feature vector. The normalized correlation metrics are used to measure the degree of similarity of two feature vectors of palmprint images. We collected 1050 palmprint images, 5 samples from each of 210 persons. Experiment results show that our proposed method can achieve an acceptable accuracy rate with FRR = 1.754, and FAR= 0.699.
This paper proposes a new technique to extract the palmprint features based on some fractal codes. The palmprint features representation is formed based on position of range blocks and direction between the position of range and domain blocks of fractal codes. Each palmprint representation is divided into a set n blocks and the mean value of each block are used to form the feature vector. The normalized correlation metrics are used to measure the degree of similarity of two feature vectors of palmprint images. We collected 1050 palmprint images, 5 samples from each of 210 persons. Experiment results show that our proposed method can achieve an acceptable accuracy rate with FRR = 1.754, and FAR= 0.699.

More info:

Published by: ijcsis on Mar 08, 2011
Copyright:Attribution Non-commercial

Availability:

Read on Scribd mobile: iPhone, iPad and Android.
download as PDF, TXT or read online from Scribd
See more
See less

03/08/2011

pdf

text

original

 
PIFS CODES BASED FORBIOMETRIC PALMPRINT VERIFICATION
 
I Ketut Gede Darma Putra
 
Departement of Electrical Engineering, Faculty of EngineeringUdayana University, Bukit Jimbaran, Bali - Indonesiaemail : duglaire@yahoo.com
 
Abstract
 This paper proposes a new technique to extractthe palmprint features based on some fractal codes. Thepalmprint features representation is formed based on positionof range blocks and direction between the position of rangeand domain blocks of fractal codes. Each palmprintrepresentation is divided into a set n blocks and the meanvalue of each block are used to form the feature vector. Thenormalized correlation metrics are used to measure thedegree of similarity of two feature vectors of palmprintimages. We collected 1050 palmprint images, 5 samples fromeach of 210 persons. Experiment results show that ourproposed method can achieve an acceptable accuracy ratewith FRR = 1.754, and FAR= 0.699.
 
Keyword 
;
biometrics, fractal codes, fractal dimension,feature extraction, palmprint recognition
I. INTRODUCTION
The personal verification becomes an important andhighly demanded technique for security access systems inthis information area. Traditional automatic personalrecognition can be divided into two categories: token-based, such as a physical key, an ID card, and a passport,and knowledge-based, such as a password and a PIN.However these approaches have some limitations. In thetoken-based approach, the “token” can be easily stolen orlost. In the knowledge-based approach, the “knowledge”can be guessed or forgotten [21]. In order to reduce thesecurity problem caused by traditional methods, biometricverification techniques have been intensively studied anddeveloped to improve reliability of personal verification.Biometric-based approach use human physiological orbehavioral features to identify a person. The most widelyused biometric features are of the fingerprints and the mostreliable are of the irises. However, it is very difficult toextract small minutiae features from unclear fingerprintsand the iris input devices are very expensive [19]. Otherbiometric features such as of face, voice, hand geometries,and handwritten are less accurate. Faces and voices can bemimicked easily, hand geometries and handwritten can befaked easily.Palmprint is the relatively new in physiologicalbiometrics [18]. There are many unique features in apalmprint image that can be used for personal recognition.Principal lines, wrinkles, ridges, minutiae points, singularpoints and texture are regarded as useful features forpalmprint representations [21]. A palmprint has severaladvantages compared to other available features: low-resolution images can be used, low cost capture devicescan be used, it is very difficult or impossible to fakepalmprints, and their characteristics are stable and unique[18].Recently, many verification/identification technologiesusing palmprint biometrics have been developed[2],[3],[4],[5],[11],[12],[13],[18],[21]. Zhang
et al
. [21]applied 2-D Gabor filter to obtain the texture features of palmprints. Pang
at al
. [13] used the pseudo-orthogonalmoments to extract the features of palmprint. LI
et al
. [12]transformed the palmprint from spatial to frequencydomain using Fourier transform and then computed ringand sector energy features. Connie at al.[2] extracted thetexture feature of palmprint using PCA and ICA. Wu
et al
.[18] extracted line feature vectors (LFV) using themagnitudes and orientations of the gradient of the pointson palm-lines. Kumar
et al
.[11] combined the palmprintsand hand geometries for verification system. Eachpalmprint was divided into overlapping blocks and thestandard deviation value of each block was used to formthe feature vector.In this paper, we propose a new technique to extract thefeatures of palmprint based on fractal codes. Thistechnique is different with the method in [4] and [5].II. IMAGE ACQUISITIONAll of palm images are captured using Sony DSC P72digital camera with resolution of 640 x 480 pixels. Eachpersons was requested to put his/her left hand palm downon with a black background. There are some pegs on theboard to control the hand oriented, translation, andstretching. A sample of the hand and pegs position on theblack board is shown on Figure 1 (a).III. PALMPRINT EXTRACTION ANDNORMALIZATIONThis paper used new technique to extract the ROI(region of interest) of palmprint. This technique consists of two steps in center of mass (centroid) method. These stepscan be explained as follow.a.
 
The gray level hand image is thresholded to obtain thebinary hand image. The threshold value was computedautomatically using the Otsu method. To avoid thewhite pixels (not pixel object) outside of the handobject is used median filter.
(IJCSIS) International Journal of Computer Science and Information Security,Vol. 9, No. 2, February 201147 http://sites.google.com/site/ijcsis/ISSN 1947-5500
 
b.
 
Each of the acquired hand images needs to be alignedin a preferred direction so as to capture the samefeatures for matching. The moment orientation methodis applied to the binary image to estimate theorientation of the hand. In the method, the angle of rotation (
θ 
) is the difference between normal axis andmajor axis of ellipse that can be computed as follows.
=
2,00,2 1,11
2tan21
µ µ µ θ 
 
(1)
 
( ) ( )
qpm nqp
nnmm
=
,
µ 
 
(2)
 
where
qp
,
µ 
represent the (p,q)
th
moment central, and(
nm
,
) represents center of area is defined as
=
m n
mm
1
,
=
m n
nn
1
, (3)where
represents number of pixel object.Furthermore, the grayscale and the binary image arerotated
 
about (
θ 
) degree.c.
 
Bounding box operation is applied to the rotatedbinary image to get the smallest rectangle whichcontains the binary hand image. The original handimage, binarized image, and the bounded imageshown in Figure 1 (a), (b), and (c), respectively.d.
 
The centroid of bounded image is computed usingequation (3) and based on this centroid, the boundedbinary and original images are segmented with 200 x200 pixels. The segmented image and its centroidposition are shown in Figure 1 (d) and (e).e.
 
The centroid of the segmented binary image iscomputed and based on this centroid the ROI of grayscale palmprint image can be cropped with size128 x 128 pixels. The first and the second positions of centroid in binary and gray level image are shown inFigure 1 (f) and (g).This method is so simple. This method has been testedfor 1050 palmprint images acquired from 210 persons, andthe results show this method is reliable.Before the feature extraction phase, the extracted ROIare normalized using normalization method in [11] toreduce the possible imperfections in the image due to non-uniform illumination. The method is as below:
otherwiseyxif  yx
φ λ φ λ φ 
>
+=
),(),('
 
(4)
 
ρ φ ρ λ 
2
}),({
=
yx
 
(5)
 
where
and
I’
represents original grayscale palmprintimage and the normalized image respectively,
φ 
 
and
 
ρ 
 
represents mean and variance of the original imagerespectively, while
φ 
and
ρ 
 
are the desired values formean and variance respectively. This research use
φ 
= 180and
ρ 
= 180 for all experiments.
(a) (b) (c)(d) (e) (f) (g)
Figure 1. Extraction of palmprint, (a) original image, (b)binary image of (a), (c) object bounded, (d) and (e)position of the first centroid mass in segmented binary andgray level image, respectively, (f) and (g) position of thesecond centroid mass in segmented binary and gray levelimage, respectively.IV. FEATURES EXTRACTIONThere are three main steps to extract the palmprintfeatures based on fractal codes proposed in this paper.These steps can be explained as follows.
A.
 
Extraction of fractal codes of palmprint images
Fractal codes of palmprint images are obtained usingthe partitioned iterated function system (PIFS) method. InPIFS method, each image is partitioned into its rangeblocks and domain blocks. The size of the domain blocksis usually larger than the size of the range blocks. Therelation between a pair of range block (
R
i
) and domainblock (
D
i
) is noted as
 
( )
iii
DwR
=
 
(6)
 
w
i
is contracted mapping that describes the similarityrelation between
R
i
and
D
i
, and is usually defined as anaffine transformation as below:
+
=
iiiiiiiiiiiiiii
oezyxscbazyxw
0000
 
(7)
 
where
x
i
and
y
i
 
represent top-left coordinate of the
R
i
, and
z
i
is the brightness value of its block. Matrix elements
a
i
,
 b
i
, c
i
,
and
i
, are the parameters of spatial rotations andflips of 
D
i
,
s
i
is the contrast scaling and
o
i
is the luminanceoffset. Vector elements
e
i
and
i
 
are offset value of space.In this paper, we used the size of domain region twice therange size, so the values of 
a
i
,
b
i
, c
i
,
and
i
are 0.5. Theactual fractal code p
i
below is usually used in practice[19].
iiiiRRDDi
ossizeyxyx
iiii
,,,,,,,
θ 
=
 
(8)
 
(IJCSIS) International Journal of Computer Science and Information Security,Vol. 9, No. 2, February 201148 http://sites.google.com/site/ijcsis/ISSN 1947-5500
 
where
ii
RR
yx
,
and
ii
DD
yx
,
represent top-leftcoordinate position of the range block and domain block,respectively, and
size
is the size of range block. The fractalcodes of a palmprint image is denoted as follow:
U
ii
1
=
=
(9)
where
represents the number of the fractal code. Theinequality expression below is used to indicate whether therange and the relevant domain block are similar or not.
( )
,,
ε 
DR
(10)
where
d(R,D)
represents rmse value, and
є
is the threshold(tolerance) value. The range and the relevant domain block is similar if 
d(R,D)
is less or equal than
є
. Otherwise, theblock is regarded not similar.
B.
 
Palmprint features representation
The first step of this method is the forming of angleimage A as follows.
( )
21
,3,2,1,,3,2,1,,
jjA
i
KK
===
α 
(11)
iRDRDi
xxyy
=
arctan
α 
 
ii
Ryand Rxjif 
==
,otherwise,0
=
i
α 
(12)where
( )
ii
DD
yx
,
represent top-left coordinate of thedomain block (see formula (8)) and
i
represent the anglebetween range and domain block. The angle image is notbinary image representation. The criterion below are addedto compute the direction
.
i
α 
 
iiDRDR
thenyyand xxif 
α 
=<
 
iiDRDR
thenyyand xxif 
α 
=>
180
 
iiDRDR
thenyyand xxif 
α 
+=>
180
 
iiDRDR
thenyyand xxif 
α 
=<
36090
==
iDRDR
thenyyand xxif 
α 
 
270
==
iDRDR
thenyyand xxif 
α 
(13)The criterion
)min(
sizesize
i
=
means the palmprintfeatures representation is formed practically using thecoordinate of the smallest size range block. Later, therepresentation is filtered as follow.
( ) ( ) ( )
,,,,
'
nxm
yxhyxyx
=
 
(14)
 
h(x,y)
is filter which all of its component are one. Figure2(b) show the palmprint features image of Figure 2(a).
 
C.
 
Palmprint feature vector 
Palmprint feature vector (
) is obtained by dividingthe palmprint image into 16 x 16 blocks, and for eachblock its mean value is computed, so obtained the featurevector
( )
vvv
,,
21
K
=
, where N = 256,and
v
i
 
ismean value of block 
i
.(a) (b)(c) (d)Figure 2. Palmprint feature extraction, (a) original image,(b) Image
, (c) Image
I’,
(d) block feature representationThe Figure 2 (d) show the palmprint feature representationin 16 x 16 sub blocks. Figure 3 shows example of threegroups of palmprints from the same palm and palms withsimilar/different line structures. The features of thesepalmprints are plotted in figure 4. The results show that thefeatures of three palm images from the same person areclose to each other than the features of three palm imagesfrom the different persons with similar or different linestructures.V. PALMPRINT FEATURE MATCHINGThe degree of similarity between two palmprintfeatures is computed as follows:
( )( )( )( )
[ ]
( )( )
[ ]
2121
1
ssss ss rs
xxxxxxxx xxxx
=
(15)where
s
xx
,
are the mean of palmprint feature
x
r
and
x
s
,respectively. The above equation computes one minusnormalized correlation between palmprint feature vector
x
r
 and
x
s
. The values of 
rs
are between 0 – 2. The
rs
willbe close to 0 if 
x
r
and
x
s
obtained from two image of thesame palmprint. Otherwise, the
rs
will be far from 0.
 
Figure 4 shows comparison of feature component of those palmprint shown in figure 3, and their score are listedin Table 1. The matching score of group A are close to 0,and the matching score of group B and C are far from 0.The average score of group A, B, and C are 0.1762,0.5057, and 0.6452, respectively. It is easy to distinguishgroup A from group B and C using these scores.(a1) (a2) (a3)
Group A: palmprints from the same person
(IJCSIS) International Journal of Computer Science and Information Security,Vol. 9, No. 2, February 201149 http://sites.google.com/site/ijcsis/ISSN 1947-5500

You're Reading a Free Preview

Download
/*********** DO NOT ALTER ANYTHING BELOW THIS LINE ! ************/ var s_code=s.t();if(s_code)document.write(s_code)//-->