Research Article Palm Vein Verification Using Multiple...

Research ArticlePalm Vein Verification Using Multiple Features and LocalityPreserving Projections

Ali Mohsin Al-juboori12 Wei Bu3 Xiangqian Wu1 and Qiushi Zhao1

1 School of Computer Science and Technology Harbin Institute of Technology Harbin 150001 China2 College of Computer Science and Mathematics University of Al-Qadisiyah Iraq3 Department of New Media Technology and Arts Harbin Institute of Technology Harbin 150001 China

Correspondence should be addressed to Xiangqian Wu xqwuhiteducn

Received 28 August 2013 Accepted 24 December 2013 Published 17 February 2014

Academic Editors J Li and X You

Copyright copy 2014 Ali Mohsin Al-juboori et al This is an open access article distributed under the Creative Commons AttributionLicense which permits unrestricted use distribution and reproduction in any medium provided the original work is properlycited

Biometrics is defined as identifying people by their physiological characteristic such as iris pattern fingerprint and face or bysome aspects of their behavior such as voice signature and gesture Considerable attention has been drawn on these issues duringthe last several decades And many biometric systems for commercial applications have been successfully developed Recentlythe vein pattern biometric becomes increasingly attractive for its uniqueness stability and noninvasiveness A vein pattern is thephysical distribution structure of the blood vessels underneath a personrsquos skin The palm vein pattern is very ganglion and it showsa huge number of vessels The attitude of the palm vein vessels stays in the same location for the whole life and its pattern isdefinitely unique In our work the matching filter method is proposed for the palm vein image enhancement New palm veinfeatures extractionmethods global feature extracted based on wavelet coefficients and locality preserving projections (WLPP) andlocal feature based on local binary pattern variance and locality preserving projections (LBPV LPP) have been proposed Finallythe nearest neighbour matching method has been proposed that verified the test palm vein images The experimental result showsthat the EER to the proposed method is 01378

1 Introduction

The hand vein pattern is here regarded as a biometric featureBiometrics is the automated measurement of physiologicalor behavioural characteristics to determine or verify identityPhysical characteristics of the human body are biological andbehavioural properties The hand vein pattern is used as abiometric features in this case a biological characteristic ofthe person can be detected and from which distinguishingThe biometric features can be extracted for the purpose ofautomated recognition of persons Verification is affirma-tion by test and provision of objective proof that specifiedrequirements have been fulfilled Verification in a biometricapplication is the outcome true or false on a claim aboutthe similarity of a biometric reference and a recognitionbiometric sample by making a comparison a one-to-onematch Biometric authentication is often used as a synonym

for verification but formally this is deprecated Biometricidentification is the outcome of a biometric system functionthat implements a one-to-many search to obtain a filter list[1] As security requirements increase biometric techniquesincluding face fingerprint iris voice and vein recognitionshave been widely used for personal identification Biometricshas been applied to building access control immigrationcontrol and user authentication for financial transactionsAlthough vein recognition has not been as widely adopted asfingerprint face and iris recognition it has some advantagesSince vein patterns exist inside the skin it is very difficult tosteal them The vein patterns are not easily altered by otherfactors such as dry or wet skin Palm vein recognition is oneformof vein pattern recognitionwhich identifies a user basedon palm vein patterns Palm vein recognition investigated inprevious studies uses vein patterns in the palm for personalidentification Hand and palm vein recognitions however

Hindawi Publishing Corporatione Scientific World JournalVolume 2014 Article ID 246083 11 pageshttpdxdoiorg1011552014246083

2 The Scientific World Journal

have a disadvantage in that the dimensions of the device areinevitably large because the vascular features from the wholehand are extracted while capturing a vein image [2]

There aremany studies on the vein pattern recognition In[2] the researchers proposed amethod for finger vein patternsusing a weighted local binary pattern (LBP) and supportvector machine (SVM) Without using any preprocessingmethods the LBP codes are extracted And classifying theLBP codes that extracted into three categories depends on theSVM classifier large amount (LA) medium amount (MA)and small amount (SA) of finger vein patterns Accordingto these areas different weights are given to the extractedLBP code The disadvantage of the work the experimentalis tested on small database contains only 960 finger veinimages In [3] the researchers worked on PolyUmultispectralpalmprint databaseThe researchers combined the palmprintand palm vein for their proposed system The matchingfilter method is used to extract the vein The EER to thesystem is 03091 based on palm vein only while theyused fused multimodal (palmprint and palm vein) featuresto evaluate the system In [4] the researchers extract mul-tiscale LBP features of hand vein images The hand veinimage is decomposed with two-level wavelet and gets eightsubband coefficient matrices and computes the weight ofeach subband the recognition accuracy of each subbandis calculated by original LBP based on Euclidean distanceThe experimental results reported the EER value to theirproposedmethod is 2067 In [5] finger vein recognition hasbeen identified using local binary pattern variance (LBPV)Global matching method is used to get more speeding andto decrease feature dimensions using distance measurementThe classification rate of this method is tested using supportvector machine (SVM) In [6] the researcher presents a mul-timodal personal identification system using palmprint andpalm vein images with their fusion applied at the image levelThe palmprint and palm vein images are fused by a new edge-preserving and contrast-enhancing wavelet fusion methodin which the modified multiscale edges of the palmprintand palm vein images are combined A palm representationcalled ldquoLaplacianpalmrdquo feature is extracted from the fusedimages by the locality preserving projections (LPP) In [7]the researcher presents two new palm vein representationsusingHessian phase information from the enhanced vascularpatterns in the normalized images and secondly from theorientation encoding of palm vein line-like patterns usinglocalizedRadon transform In [8] the researchers consider thepalm vein as a texture and apply feature extraction based ontexture vein for person authentication The feature extractedfrom the palm vein is based on 2D Gabor filter Then adirectional code technique is proposed to represent the palmvein features called vein code The matching between twovein codes is computed by normalized Hamming distance In[9] the researchers worked on PolyUmultispectral palmprintdatabase They used a multiscale curvelet transform as afeature extraction and used a subset from the features formatching using Hamming distance The lowest EER to theirsystem is 066 which is got when selected (40) from thefeatures set In [10] the vein feature representation methodis called orientation of local binary pattern (OLBP) which is

an extension of local binary pattern (LBP) Based on OLBPfeature representation construct a hand vein recognitionsystem employing multiple hand vein patterns that includepalm vein dorsal vein and three finger veins (index middleand ring finger) Vein images are enhanced using Gaussianmatched filter and extracted OLBP features and matchedFinally the matching scores are fused using support vectormachine (SVM) to make a decision In the previous work[11] we implemented a palm vein verification system Thefeatures extraction depends on the Gabor filter with 8 scalesand 8 directions A new dimension reduction method isproposed called Fisher Vein The matching step is dependingon the Nearest Neighbour method The EER to the system is02335

This paper proposes a multiple features extraction basedon global and locate features and merging with localitypreserving projections (LPP) The global features extractedusing wavelet transform coefficients combine with local-ity preserving projections and create feature vector calledwavelet locality preserving projections (WLPP) and the localbinary pattern variance (LBPV) represents the locale featuresfor the palm vein image combined with locality preservingprojections and create feature vector called local binary pat-tern variance-locality preserving projections (LBPV-LPP)Based on the proposed palm vein features representationmethods a palm vein authentication system is constructedThe nearest neighbour method is proposed to match thetest palm vein images The system flowchart is shown inFigure 1 First of all the palm vein image is enhanced usingmatching filter and then the features are extracted (WLPP andLBPV-LPP) After similarity measure for each feature fusiontechnique is applied to fuse all the matching scores to obtaina final decision

The rest of this paper is organized as follows Section 2describes the preprocessing Section 3 describes the featureextraction methods Section 4 describes the matching andmatching score fusion strategy used in this work Finally theexperimental result and conclusions are drawn in Section 5

2 Preprocessing

It is observed that the cross-sections of palm veins aresimilar to Gaussian functions Based on this observation thematched filters [14 15] which are widely used in retinal vesselextraction can be a good technique to extract these palmveins The matched filters are Gaussian shaped filters alongangle 0

Consider

119892120590

120579(119909 119910) = minus exp(minus

11990910158402

1205902) minus 119898

for 10038161003816100381610038161003816119909101584010038161003816100381610038161003816le 3120590

10038161003816100381610038161003816119910101584010038161003816100381610038161003816le

119871

2

(1)

where 1199091015840

= 119909 cos 120579 + 119910 sin 120579 1199101015840

= minus119909 sin 120579 + 119910 cos 120579 120590 is thestandard deviation of Gaussian m is the mean value of thefilter and 119871 is the length of the filter in 119910 direction which isset empirically In order to suppress the background pixelsthe filter is designed as a zero-sum For one 120590 six different

The Scientific World Journal 3

Palm veinimage

Palm veinimage

Matchingenhancement

filter

Matchingenhancement

filter

Waveletfeature

extraction

Wavelet Waveletfeature

extraction

LBPVfeature

extraction

LBPVfeature

extraction

LPP

LPP LPP

WLPP

WLPP matchingscore

matchingscore

LBPV

Feature extraction

Feature extraction

Fusion

Matching and matching fusion

Database

Enrollment steps

Decision

Projection

Projection

Projection

Projection

Verify steps

LBPV

LPPLBPV

Figure 1 System flowchart

angle filters (120579119895 = 1198951205876 where 119895 = 0 1 2 3 4 5) are appliedfor each pixel and the maximal response among these sixdirections is kept as the final response for the given scale

119877120590

119865= max (119877

120590

120579119895(119909 119910)) 119895 = 0 1 2 3 4 5

119877120590

120579119895(119909 119910) = 119892

120590

120579119895lowast 119891 (119909 119910)

(2)

where 119891(119909 119910) is the original image and lowast denotes the con-volution operation [3 14 15] The value of 120590 is set empiricalFigure 2 shows the result of palm vein enhancement

3 Feature Extraction

The feature extraction step aims to extract the actual featuresof the palm vein pattern from the image and then used it forthe matching If the image is an enrolled sample the featuresare saved in a training database for later matching Once thefeatures are extracted they are compared with the ones in thedatabase and based on that comparison a decision is taken Inthe proposework newpalm vein features extractionmethodsare proposed based onwavelet locality preserving projections(WLPP) features and local binary pattern variance-localitypreserving projections (LBPV LPP) features

31 Feature Extraction Based on 2D Wavelet Transform Areliable and robust feature extraction is important for patternrecognition tasks Recently a lot of research work has beendirected towards using wavelet based features The discretewavelet transform (DWT) has a good time and frequency res-olution and hence it can be used for extracting the localizedcontributions of the signal of interest The mathematical toolfor a hierarchically decomposing is the wavelet transformThe wavelet function described the coarse overall shape anddetails that range from wide to close The wavelet functionan efficient technique for representing the image surface

and curve The discrete wavelet transform (DWT) is a verypopular tool for the analysis of nonstationary signalsThe ideaof it is to represent a signal as a series of approximations low-pass version corresponds to the signal and high-pass versioncorresponds to details This is done at different resolutionsIt is almost equivalent to filtering the signal with a bank ofband pass filters whose impulse responses are all roughlygiven by scaled versions of a mother wavelet [4 16 17]Multidimensional wavelets (especially the two dimensional2Dwavelets) can be treated as two 1Dwavelet transforms (one1D wavelet transform along the row direction and the other1D wavelet transform along the column direction) Thus the2Dwavelet transform can be computed in cascade by filteringthe rows and columns of images with 1D filters Figure 3shows a one-octave decomposition of an image into fourcomponents low-pass rows low-pass columns (LL) low-pass rows high-pass columns (LH) high-pass rows low-passcolumns (HL) high-pass rows high-pass columns (HH)Thecoefficients obtained by applying the 2Dwavelet transformonan image are called the subimages of wavelet transform [12]Figure 4 shows the result of db2 on the enhanced palm veinimage

32 Feature Extraction Based on Local Binary Pattern Vari-ance (LBPV) The local binary pattern variance (LBPV) com-putes the variance (VAR) from a local region and accumulatesit into the local binary pattern bin This can be considered asthe integral projection along the VAR coordinate The spatialstructure of the local image texture can be represented by LBPThe pattern number is computed by comparing the centerpixel value with those of its neighbours as shown in Figure 5The LBP is computed by

LBP119875119877

=

119875minus1

sum

119901=0

119904 (119892119875

minus 119892119888) 2119875


Original image

048 05205 054 056 058 06206 064 0660

20

40

60

80Original image histogram

(a)

Enhancement image

01 02 03 04 05 06 07 08 090 10

20

40

60

80

100

120Enhancement image histogram

(b)

Figure 2 Palm vein image enhancements

Image

LL

LL

LH

HL

HH

Filter in x-direction Filter in y-direction

H

L

HH

g(x y)

2

2

2

2

2

2

Figure 3 Decomposition algorithm of 2D wavelet transform [12]

Figure 4 db2 result


(P = 4 R = 10) (P = 8 R = 10) (P = 12 R = 15)

g1

g2

g3

g0gc

Figure 5 LBP neighbours sets for different (119875 119877) [5 13]

(a)

0 20 40 600

2

4

6

8

10

12

14

16

18

20

(b)

Figure 6 Palm vein image and LBPV histogram

119904 (119909) = 1 119909 ge 0

0 119909 lt 0(3)

where 119892119888is the gray value of the central pixel 119892

119875is the

value of its neighbours P is the number of neighboursand 119877 is the radius of the neighbourhoods Suppose thatthe coordinates of 119892

119888are (00) then the coordinates of 119892

119875

are given by (minus119877 sin(2120587119901119875) 119877 cos(2120587119901119875)) Figure 4 showsexamples of circularly symmetric neighbour sets for differentconfigurations of (119875 119877) By using the interpolation the grayvalue of the neighbors can be estimated [5 13 18]

The local spatial pattern and local contrast can bedescribed briefly by LBPPRVARPR VARPR has a continuousvalue that has to be quantized which can be done by firstcalculating feature distributions from all training images toget a total distribution and then to guarantee the highestquantization resolution some threshold values are computedto partition the total distribution into 119873 bins with an equalnumber of entries The threshold values are used to quantizethe VAR of the test images There are three particularlimitations to this quantization method as stated by [13] The

LBPVproposes a solution to these problems of the descriptorThe information of the variance VARPR is not involved in thecalculation of the LBP histogram The histogram operationallocates the same weight to each LBP pattern independentof the LBP variance of the local region The LBPV is given as

LBPV119875119877

(119870) =

119873

sum

119894=1

119872

sum

119895=1

119882 (LBP119875119877

(119894 119895) 119896) 119896 isin [0 119870]

119882 (LBP119875119877

(119894 119895) 119896) = VAR119875119877

(119894 119895) LBP119875119877

(119894 119895) = 119896

0 otherwise(4)

where 119870 is the maximal LBP pattern value It is observed thatLBPV is a simplified descriptorwhose feature size is small thatcan be utilized in many applications It is also training-freeand it does not need quantization Figure 6 shows the resultof the LBPV histogram for the palm vein image

33 Locality Preserving Projections In most computer visionand pattern recognition problems the large number of


sensory inputs such as images and videos is computationallychallenging to analyze In such cases it is desirable to reducethe dimensionality of the data while preserving the originalinformation in the data distribution allowing for moreefficient learning and inference If the variance of the mul-tivariate data is faithfully represented as a set of parametersthe data can be considered as a set of geometrically relatedpoints lying on a smooth low-dimensional manifold Thefundamental issue in dimensionality reduction is how tomodel the geometry structure of the manifold and produce afaithful embedding for data projection Linear dimensionalityreduction (LDR) techniques have been increasingly impor-tant in pattern recognition since they permit a relativelysimple mapping of data onto a lower dimensional subspaceleading to simple and computationally efficient classificationstrategiesThemain advantage of the linear methods over thenonlinear ones is that the embedding function of the lineartechniques is defined everywhere in the input space while fornonlinear embedding techniques it is only defined for a setof data samples [19] Locality preserving projection (LPP) is amanifold learningmethodwidely used in pattern recognitionand computer vision LPP is also well known as a linear graphembedding methodWhen LPP transforms different samplesinto new representations using the same linear transform ittries to preserve the local structure of the samples that is theneighbor relationship between samples so that samples thatwere originally in close proximity in the original space remainso in the new space [20]

The locality preserving projections (LPP) are linear pro-jective maps that arise by solving a variational problemthat optimally preserves the neighborhood structure of thedata set LPP should be seen as an alternative to principalcomponent analysis (PCA) which is a classical linear tech-nique that projects the data along the directions of maximalvariance When the high-dimensional data lies on a low-dimensional manifold embedded in the ambient space thelocality preserving projections are obtained by finding theoptimal linear approximations to the eigenfunctions of theLaplace Beltrami operator on the manifold The LPP buildsa graph incorporating neighborhood information of the dataset Using the notion of the Laplacian of the graph computea transformation matrix which maps the data points toa subspace This linear transformation optimally preserveslocal neighborhood information in a certain sense The rep-resentation map generated by the algorithm may be viewedas a linear discrete approximation to a continuous map thatnaturally arises from the geometry of the manifold localitypreserving projection (LPP) which is a linear approximationof the nonlinear Laplacian Eigenmap In the proposed workthe algorithm procedure that is used as show in [21ndash23]

(1) Constructing the Adjacency Graph Let 119866 denote a graphwith119898 nodesWe put an edge between nodes 119894 and 119895 if 119909

119894and

119909119895are ldquocloserdquo There are two variations

(a) 120598-neighborhoods (parameter 120598 isin 119877) nodes 119894 and 119895

are connected by an edge if 119909119894minus 1199091198952

lt 120598 where thenorm is the usual Euclidean norm in 119877

119899

(b) K nearest neighbors (parameter 119896 isin 119873) nodes 119894 and119895 are connected by an edge if 119894 is among 119896 nearestneighbors of 119895 or 119895 among 119896 nearest neighbors of i

(2)Choosing theWeights Here as well we have two variationsfor weighting the edges 119882 is a sparse symmetric 119898 times

119898 matrix with 119882119894119895having the weight of the edge joining

vertices 119894 and 119895 and 0 if there is no such edge

(a) Heat kernel (Parameter 119905 isin 119877) if nodes 119894 and 119895 areconnected put

119882119894119895

= 119890minus119909119894minus119909119895

2119905

(5)

(b) Simple-mined (no parameter) 119882119894119895

= 1 if and only ifvertices 119894 and 119895 are connected by an edge

(3) Eigenmaps Compute the eigenvectors and eigenvalues forthe generalized eigenvector problem

119883119871119883119879

119886 = 120582119883119863119863119879

119886 (6)

where 119863 is a diagonal matrix whose entries are column (orrow since 119882 is symmetric) sums of 119882 119863

119894119894= sum 119895119882

119895119894 119871 =

119863 minus 119882 is the Laplacian matrixThe 119894th column of matrix 119883 is119909119894Let the column vectors 119886

0 119886

119894minus1be the solutions of (6)

ordered according to their eigenvlaues 1205820

lt sdot sdot sdot lt 120582119897minus1

Thusthe embedding is as follows

119909119894997888rarr 119910119894= 119860119879

119909119894 119860 = (119886

0 1198861 119886

119897minus1) (7)

where 119910119894is a l-dimensional vector and 119860 is a n times lmatrix

In the proposed work two types of features vectors arecreated (global and local) The global features extracted arebased on wavelet and locality preserving projections (LPP)Two types of the wavelet functions are used (Daubechies(db2) and Symlets (sym2)) For each enhanced palm veinimage calculate the above functions of wavelet The thirdlevels for each earlier wavelet functions are computed andselected the approximation coefficients only as palm veinfeatures Then do the normalization for db2 and sym2approximations coefficients based on 119885-normalization Thetwo normalized feature vectors are fusions based on concate-nation to create a new feature vector The modern featurevector is projected based on locality preserving projectionsfrom high-dimension space to low-dimension space Thatfeatures vector is called wavelet locality preserving projec-tions (WLPP)

The local features extracted based on local binary patternvariance (LBPV) and locality preserving projections (LPP)The LBPV is computed to the enhanced palm vein imagesand also to the approximationrsquos coefficients of the waveletfunctions Daubechies (db2) and Symlets (sym2) That pro-cess gave three feature vectors one from enhanced imageand the others from the wavelet approximationrsquos coefficientThese feature vectors are normalized using 119885-normalizationThe normalization vectors are fused based on concatena-tion Based on locality preserving projections project the


Palm veinimage (I)

sym2(I)

db2(I)

Apx(sym2(I))

Apx(db2(I))

LBPVLBPV(Apx(sym2))

LBPV(I)

LBPV(Apx(db2))

Z-normalization

Z-normalization

Z-normalization

Z-normalization

Z-normalization

Fusio

nFu

sion

LPP WLPP

LPP

Projection

Projection LBPV LPP

Figure 7 Flowchart of the feature vectors extraction

concatenated featurersquos vector from high-dimension spaceto low-dimension space That features vector is calledlocal binary pattern variance locality preserving projections(LBPV LPP) Figure 7 shows the flowchart of creation thefeature vectors

4 Matching and Matching Score Fusion

The nearest neighbor method is used to compute the match-ing between the train set and test set To measure thesimilarity between two biometric feature vectors we usedEuclidean distance as similarity measures Let 119910 denote thetest feature vector and 119909

119896

119894119894 = 1 sdot sdot sdot 119862

119896119896 = 1 sdot sdot sdot 119862 represent

the 119894th gallery image of subject ID119896 where 119862

119896is the number

of images of subject ID119896and 119862 is the totally numbers of the

images in the train set The smallest Euclidean distance [24]is

ID119910

= argmin119896

10038171003817100381710038171003817119910 minus 119909119896

119894

10038171003817100381710038171003817

2

(8)

In the proposed system we have two feature vectors(WLPP and LBPV LPP) by using the nearest neighbourcompute thematching distancematrix to each features vector(WLPP distance matrix and LBPV LPP distance matrix) Inthis paper weighted sum fusion rule is applied to fusing thetwo matching distance matrices to obtain a final matchingscore to make a decision as shown in Figure 1 If the optimalweights of two modalities are searched exhaustively it needsa twofold loop to reach to smallest value of EER

5 Experimental Result and Conclusion

51 Experimental Result Experiments have been performedto evaluate the effectiveness of proposed palm vein veri-fication methods based on multispectral palmprint PolyUdatabase The biometric research centre at the Hong KongPolytechnic University has developed a real time multispec-tral palmprint capture device that can capture palmprintimages under blue green red and near-infrared (NIR)illuminations and has been used to construct a large-scalemultispectral palmprint database The database contains

Table 1 Experimental result without image enhancement

Matching method EEREuclidian matching 04899Manhattan matching 05274Cosine matching 01888Correlation matching 01944

6000 images from 500 different palms for each of the aboveillumination The proposed method uses the near-infrared(NIR) illumination images of PolyU multispectral palmprintdatabase [25]

Each individual has 12 palm vein images six palm veinimages are used for enrollment and the other six imagesare used for the testing of each individual in the experi-ments We used two performance measures namely the falserejection rate (FAR) and the false acceptance rate (FRR) Forcomputing FRR value we compare the biometric referencewith all samples of the same individual For computing FARvalue we compare the biometric reference of an individualwith all samples from different individuals The disparitiesdistribution between intraclass and interclass matching resulthave been plotted It shows the separation between genuineand impostors The receiver operating characteristic (ROC)curve is between FRR and FAR Equal error rate (EER)is the point where FRR is equal to FAR and the smallerEER indicates a better performance The nearest neighbourmethod is used to verify the feature vector from test set withthe train set feature vectors and take the minimum distancefor verification The proposed system has many steps asimage enhancement and the two types of features extractionmethods and the matching classifier method The followingexperimental results show when execute the system withoutimage enhancement or without using dimension reductionor based on one type of feature extraction Figure 8(a) showsthe distribution between genuine user and impostor to thesmallest EER value classifier without image enhancementFigure 8(b) shows the ROC curve when implement thesystem without image enhancement with different classifierand Table 1 shows the EER value with different classifiermethods


0 02 04 06 08 1 12 14Distance

Genuine userImpostor

0123456789

Freq

uenc

e (

)

(a)

FAR ()

Euclidian distanceCosine distance

Correlation distanceManhattan distance

10minus4

10minus3

10minus2

10minus1

100

0

05

1

15

2

25

3

FRR

()

(b)

Figure 8 (a)Distribution of the genuine user and impostorwithout enhancement using cosine classifier (b) ROCcurvewithout enhancementto different classifiers

0 02 04 06 08 1 12 140

051

152

253

35

Distance

Freq

uenc

e (

)


(a)

FAR ()

Euclidian distanceCosine and correlation distanceManhattan distance

10minus5

10minus4

10minus3

10minus2

10minus1

100

101

102

20

468

1012141618

FRR

()

(b)

Figure 9 (a) Distribution of the genuine user and impostor without LPP using Euclidian classifier (b) ROC curve without LPP to differentclassifiers

Figures 8(a) and 8(b) and Table 1 show that the resultswithout image enhancement are almost acceptable becausethe proposed system uses two types of feature extraction localand global (WLPP and LBPV LPP) and also the fusion ruleis applied to fuse the two matching distance matrices (WLPPdistance matrix and LBPV LPP distance matrix) to obtain afinal matching score We get the best result while using theCosinematching classifier Figure 9(a) shows the distributionbetween genuine user and impostor to the smallest EERvalue classifier without LPP Figure 9(b) shows the ROCcurve when implement the systemwithout LPPwith differentclassifiers and Table 2 shows the EER value with differentclassifier methods

Figures 9(a) and 9(b) and Table 2 show that the resultswithout LPP are not acceptable because in the proposedsystem the feature extraction methods (Wavelet and LBPV)feature vectors have some redundancy values and appliedto fusion of the two matching distance matrices (Waveletdistance matrix and LBPV distance matrix) to obtain a finalmatching scoreWe get the best result while using the Euclid-ian matching classifier Figure 10(a) shows the distribution

Table 2 Experimental result without LPP


between genuine user and impostor to the smallest EER valueclassifier using WLPP Figure 10(b) shows the ROC curvewhen implementing the system based on WLPP featuresvector with different classifier and Table 3 shows the EERvalue with different classifier methods

Figures 10(a) and 10(b) and Table 3 show that the resultsbased onWLPP features vector are almost acceptable becausewavelet feature vector give useful information on the veinfeatures as global feature vector and the LPP method removeall the redundancy features Only wavelet distance matrixis used to obtain a final matching score The best resultis obtained while using the Euclidian matching classifier


0 02 04 06 08 1 12 140

051

152

253

35

454

Distance

Freq

uenc

e (

)


(a)

FAR ()



10minus5

10minus4

10minus3

10minus2

10minus1

100

101

102

02468

101214

FRR

()

(b)

Figure 10 (a) Distribution of the genuine user and impostor basedWLPP using Euclidian classifier (b) ROC curve basedWLPP to differentclassifiers

0 02 04 06 08 1 12 140

051

152

253

354

Distance

Freq

uenc

e (

)


(a)

FAR ()



10minus5

10minus4

10minus3

10minus2

10minus1

100

101

102

0102030405060708090

100

FRR

()

(b)

Figure 11 (a) Distribution of the genuine user and impostor based LBPV LPP using Manhattan classifier (b) ROC curve based LBPV LPPto different classifiers

Table 3 Experimental result based WLPP features vector


Figure 11(a) shows the distribution between genuine userand impostor to the smallest EER value classifier usingLBPV LPP Figure 11(b) shows the ROC curve when imple-menting the system based on LBPV LPP features vector withdifferent classifier and Table 4 shows the EER value withdifferent classifier methods

Figures 11(a) and 11(b) and Table 4 show that theresults based on LBPV LPP are not acceptable because theLBPV LPP get the local palm vein feature that is not enoughfor verification and only uses LBPV LPP distance matrix toobtain a final matching score We get the best result whileusing the Manhattan matching classifier Figure 12(a) showsthe distribution between genuine user and impostor to the

Table 4 Experimental result LBPV LPP


smallest EER value classifier for the proposed system Fig-ure 12(b) shows the ROC curve when implementing the sys-tem based on image enhancement andWLPP and LBPV LPPfeatures vector with different classifiers and Table 5 shows theEER value with different classifier methods

The distance distribution of genuine and impostor of thepalm vein images is shown in Figure 12(a) and the ROCcurve in Figure 12(b) The EER to the proposed systemis 01378 by using nearest neighbor (Euclidian distance)method This result is more acceptable from the other aboveexperimental results because the image enhancement canremove all noise and make the images vein clearer WLPPextracted the global features to the vein images and when


0 02 04 06 08 1 12 140

051

152

253

354

Distance

Freq

uenc

e (

)


(a)

FAR ()



10minus4

10minus3

10minus2

10minus1

100

002040608

112141618

2

FRR

()

(b)

Figure 12 (a) Distribution of the genuine user and impostor to the proposed system using Euclidian classifier (b) ROC curve to differentclassifiers

Table 5 Experimental result proposed system


combined with LBPV LPP local features give the best resultin verification

Five methods for palm vein authentication are proposedfor comparison In the [3 9 11] the researchers worked onthe PolyU database and we implement the methods used in[8 10] for testing on the PolyU database Table 6 shows thecomparison of our method and all the above methods

From the results illustrated in Table 6 and Figures 12(a)and 12(b) it is verified that the proposed method has betterperformance from the methods that are described in [3 8ndash11] and the lowest EER value to the proposed method is01378 by using the Euclidian distance The main benefit ofthe proposed method is that which uses the combination ofthe local and global feature vectors (WLPP and LBPV LPP)and uses the matching score fusion method that reaches tolowest EER value

52 Conclusion This paper has addressed the problems ofpalm vein segmentation and verification The matching filteris exploited to extract palm vein pattern Then global andlocal features are used (WLPP and LBPV LPP) Finally thepalm vein verification was implemented using Euclidiandistance classifier and then weighted sum fusion rule isapplied to combine the two matching distance matricesBy the proposed method we get a lower EER value equalto 01378 We can conclude that the preprocessing stepis important because the palm vein images are noisy andunclear In addition one type of the features vector (local orglobal) features is not enough for verification and the LPPmethod removes all the redundancy in the features vector

Table 6 The EER comparisons with other methods

Method EERZhang et al [3] 03091Lee [8] 1111Sun and Abdulla [9] 066Bu et al [10] 01559Al-juboori et al [11] 02335Proposed method using Euclidian distance 01378

The experimental result shows the best performance whenenhancing the vein images and use a fusion of two types of thefeature vectors This process gives us an accurate and robustpersonal verification system

Conflict of Interests

The authors declare that there is no conflict of interestsregarding the publication of this paper

Acknowledgments

This work was supported by the Natural Science Foundationof China (Grants no 61073125 and 61350004) and the Funda-mental Research Funds for the Central Universities (GrantsnoHITNSRIF2013091 and HITHSS201407)

References

[1] Annemarie NadortThe Hand Vein Pattern Used as a BiometricFeature Master Literature Thesis Amsterdam The Nether-lands 2007

[2] H C Lee B J Kang E C Lee and K R Park ldquoFinger veinrecognition using weighted local binary pattern code based ona support vector machinerdquo Journal of Zhejiang University C vol11 no 7 pp 514ndash524 2010

[3] D Zhang Z Guo G Lu L Zhang Y Liu andW Zuo ldquoOnlinejoint palmprint and palmvein verificationrdquo Expert Systems withApplications vol 38 no 3 pp 2621ndash2631 2011


[4] Y-D Wang Q-Y Yan and K-F Li ldquoHand vein recognitionbased on multi-scale LBP and waveletrdquo in Proceedings ofthe International Conference on Wavelet Analysis and PatternRecognition (ICWAPR rsquo11) pp 214ndash218 Guilin China July 2011

[5] K -Q Wang A S Krisa X -Q Wu and Q -S Zhao ldquoFingervein recognition using LBP variance with global matchingrdquo inProceedings of the International Conference on Wavelet Analysisand Pattern Recognition pp 196ndash200 2012

[6] J-G Wang W-Y Yau A Suwandy and E Sung ldquoPersonrecognition by fusing palmprint and palm vein images basedon ldquoLaplacianpalmrdquo representationrdquo Pattern Recognition vol41 no 5 pp 1514ndash1527 2008

[7] Y Zhou and A Kumar ldquoContactless palm vein identificationusing multiple representationsrdquo in Proceedings of the 4th IEEEInternational Conference onBiometricsTheory Applications andSystems (BTAS rsquo10) September 2010

[8] J -C Lee ldquoA novel biometric system based on palm vein imagerdquoPattern Recognition Letters vol 33 pp 1520ndash1528 2012

[9] J Sun and W Abdulla ldquoPalm vein recognition using curvelettransformrdquo in Proceedings of the 27th Conference on Image andVision Computing New Zealand pp 435ndash439 2012

[10] W Bu X Wu and E Gao ldquoHand vein recognition based onorientation of LBPrdquo in The International Society for OpticalEngineering vol 8371 of Proceedings of SPIE 2012

[11] A M Al-juboori W Bu X Wu and Q Zhao ldquoPalm vein ver-ification using gabor filterrdquo International Journal of ComputerScience Issues vol 10 no 1 pp 678ndash684 2013

[12] G Lu K Wang and D Zhang ldquoWavelet based feature extrac-tion for palmprint identificationrdquo in Proceedings of the 2ndInternational Conference on Image and Graphics pp 780ndash784August 2002

[13] Z Guo L Zhang and D Zhang ldquoRotation invariant textureclassification using LBP variance (LBPV)with globalmatchingrdquoPattern Recognition vol 43 no 3 pp 706ndash719 2010

[14] A Hoover V Kouznetsova andM Goldbaum ldquoLocating bloodvessels in retinal images by piecewise threshold probing of amatched filter responserdquo IEEETransactions onMedical Imagingvol 19 no 3 pp 203ndash210 2000

[15] Y B Zhang Q Li J You and P Bhattacharya ldquoPalm veinextraction and matching for personal authenticationrdquo in Pro-ceedings of the 9th International Conference on Advances inVisual Information Systems (VISUAL rsquo07) pp 154ndash164 2007

[16] M M M Fahmy ldquoPalmprint recognition based on Mel fre-quency Cepstral coefficients feature extractionrdquo Ain ShamsEngineering Journal vol 1 no 1 pp 39ndash47 2010

[17] E J Stollnitz T D DeRose and D H Salestin ldquoWavelets forcomputer graphics a primer part 1rdquo IEEE Computer Graphicsand Applications vol 15 no 3 pp 76ndash84 1995

[18] S R Ahirrao and D S Bormane ldquoA novel approach for facerecognition using local binary patternrdquo International Journal ofImage Processing andVision Sciences vol 1 no 1 pp 25ndash31 2012

[19] F Dornaika and A Assoum ldquoEnhanced and parameterlesslocality preserving projections for face recognitionrdquoNeurocom-puting vol 99 pp 448ndash457 2013

[20] Y Xu A Zhong J Yang and D Zhang ldquoLPP solution schemesfor use with face recognitionrdquo Pattern Recognition vol 43 no12 pp 4165ndash4176 2010

[21] X He and P Niyogi ldquoLocality preserving projectionsrdquo inProceedings of the Conference onAdvances inNeural InformationProcessing Systems 2003

[22] D Cai X He and J Han ldquoDocument clustering using localitypreserving indexingrdquo IEEE Transactions on Knowledge andData Engineering vol 17 no 12 pp 1624ndash1637 2005

[23] D Cai X He and J Han ldquoUsing graphmodel for face analysisrdquoTech Rep UIUCDCS-R-2005-2636 University of Illinois atUrbana-Champaign 2005

[24] M T Ibrahim Y Wang L Guan and A N VenetsanopoulosldquoA filter bank based approach for rotation invariant fingerprintrecognitionrdquo Journal of Signal Processing Systems vol 68 pp401ndash414 2012

[25] ldquoMultispectral PolyU databaserdquo httpwww4comppolyueduhksimbiometrics

Submit your manuscripts athttpwwwhindawicom

Computer Games Technology

International Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014


Distributed Sensor Networks


Advances in

FuzzySystems

Hindawi Publishing Corporationhttpwwwhindawicom

Volume 2014


ReconfigurableComputing

Hindawi Publishing Corporation httpwwwhindawicom Volume 2014


Applied Computational Intelligence and Soft Computing

thinspAdvancesthinspinthinsp

Artificial Intelligence

HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014

Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014


Electrical and Computer Engineering

Journal of

Journal of

Computer Networks and Communications


Hindawi Publishing Corporation

httpwwwhindawicom Volume 2014

Advances in

Multimedia


Biomedical Imaging


ArtificialNeural Systems

Advances in


RoboticsJournal of



Computational Intelligence and Neuroscience

Industrial EngineeringJournal of


Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014


Human-ComputerInteraction

Advances in

Computer EngineeringAdvances in



have a disadvantage in that the dimensions of the device areinevitably large because the vascular features from the wholehand are extracted while capturing a vein image [2]

There aremany studies on the vein pattern recognition In[2] the researchers proposed amethod for finger vein patternsusing a weighted local binary pattern (LBP) and supportvector machine (SVM) Without using any preprocessingmethods the LBP codes are extracted And classifying theLBP codes that extracted into three categories depends on theSVM classifier large amount (LA) medium amount (MA)and small amount (SA) of finger vein patterns Accordingto these areas different weights are given to the extractedLBP code The disadvantage of the work the experimentalis tested on small database contains only 960 finger veinimages In [3] the researchers worked on PolyUmultispectralpalmprint databaseThe researchers combined the palmprintand palm vein for their proposed system The matchingfilter method is used to extract the vein The EER to thesystem is 03091 based on palm vein only while theyused fused multimodal (palmprint and palm vein) featuresto evaluate the system In [4] the researchers extract mul-tiscale LBP features of hand vein images The hand veinimage is decomposed with two-level wavelet and gets eightsubband coefficient matrices and computes the weight ofeach subband the recognition accuracy of each subbandis calculated by original LBP based on Euclidean distanceThe experimental results reported the EER value to theirproposedmethod is 2067 In [5] finger vein recognition hasbeen identified using local binary pattern variance (LBPV)Global matching method is used to get more speeding andto decrease feature dimensions using distance measurementThe classification rate of this method is tested using supportvector machine (SVM) In [6] the researcher presents a mul-timodal personal identification system using palmprint andpalm vein images with their fusion applied at the image levelThe palmprint and palm vein images are fused by a new edge-preserving and contrast-enhancing wavelet fusion methodin which the modified multiscale edges of the palmprintand palm vein images are combined A palm representationcalled ldquoLaplacianpalmrdquo feature is extracted from the fusedimages by the locality preserving projections (LPP) In [7]the researcher presents two new palm vein representationsusingHessian phase information from the enhanced vascularpatterns in the normalized images and secondly from theorientation encoding of palm vein line-like patterns usinglocalizedRadon transform In [8] the researchers consider thepalm vein as a texture and apply feature extraction based ontexture vein for person authentication The feature extractedfrom the palm vein is based on 2D Gabor filter Then adirectional code technique is proposed to represent the palmvein features called vein code The matching between twovein codes is computed by normalized Hamming distance In[9] the researchers worked on PolyUmultispectral palmprintdatabase They used a multiscale curvelet transform as afeature extraction and used a subset from the features formatching using Hamming distance The lowest EER to theirsystem is 066 which is got when selected (40) from thefeatures set In [10] the vein feature representation methodis called orientation of local binary pattern (OLBP) which is

an extension of local binary pattern (LBP) Based on OLBPfeature representation construct a hand vein recognitionsystem employing multiple hand vein patterns that includepalm vein dorsal vein and three finger veins (index middleand ring finger) Vein images are enhanced using Gaussianmatched filter and extracted OLBP features and matchedFinally the matching scores are fused using support vectormachine (SVM) to make a decision In the previous work[11] we implemented a palm vein verification system Thefeatures extraction depends on the Gabor filter with 8 scalesand 8 directions A new dimension reduction method isproposed called Fisher Vein The matching step is dependingon the Nearest Neighbour method The EER to the system is02335

This paper proposes a multiple features extraction basedon global and locate features and merging with localitypreserving projections (LPP) The global features extractedusing wavelet transform coefficients combine with local-ity preserving projections and create feature vector calledwavelet locality preserving projections (WLPP) and the localbinary pattern variance (LBPV) represents the locale featuresfor the palm vein image combined with locality preservingprojections and create feature vector called local binary pat-tern variance-locality preserving projections (LBPV-LPP)Based on the proposed palm vein features representationmethods a palm vein authentication system is constructedThe nearest neighbour method is proposed to match thetest palm vein images The system flowchart is shown inFigure 1 First of all the palm vein image is enhanced usingmatching filter and then the features are extracted (WLPP andLBPV-LPP) After similarity measure for each feature fusiontechnique is applied to fuse all the matching scores to obtaina final decision

The rest of this paper is organized as follows Section 2describes the preprocessing Section 3 describes the featureextraction methods Section 4 describes the matching andmatching score fusion strategy used in this work Finally theexperimental result and conclusions are drawn in Section 5

2 Preprocessing

It is observed that the cross-sections of palm veins aresimilar to Gaussian functions Based on this observation thematched filters [14 15] which are widely used in retinal vesselextraction can be a good technique to extract these palmveins The matched filters are Gaussian shaped filters alongangle 0

Consider

119892120590

120579(119909 119910) = minus exp(minus

11990910158402

1205902) minus 119898

for 10038161003816100381610038161003816119909101584010038161003816100381610038161003816le 3120590

10038161003816100381610038161003816119910101584010038161003816100381610038161003816le

119871

2

(1)

where 1199091015840

= 119909 cos 120579 + 119910 sin 120579 1199101015840

= minus119909 sin 120579 + 119910 cos 120579 120590 is thestandard deviation of Gaussian m is the mean value of thefilter and 119871 is the length of the filter in 119910 direction which isset empirically In order to suppress the background pixelsthe filter is designed as a zero-sum For one 120590 six different


Palm veinimage

Palm veinimage

Matchingenhancement

filter

Matchingenhancement

filter

Waveletfeature

extraction


extraction

LBPVfeature

extraction

LBPVfeature

extraction

LPP

LPP LPP

WLPP

WLPP matchingscore

matchingscore

LBPV

Feature extraction

Feature extraction

Fusion


Database

Enrollment steps

Decision

Projection

Projection

Projection

Projection

Verify steps

LBPV

LPPLBPV



119877120590

119865= max (119877

120590

120579119895(119909 119910)) 119895 = 0 1 2 3 4 5

119877120590

120579119895(119909 119910) = 119892

120590

120579119895lowast 119891 (119909 119910)

(2)







LBP119875119877

=

119875minus1

sum

119901=0

119904 (119892119875

minus 119892119888) 2119875


Original image

048 05205 054 056 058 06206 064 0660

20

40

60


(a)

Enhancement image

01 02 03 04 05 06 07 08 090 10

20

40

60

80

100


(b)


Image

LL

LL

LH

HL

HH


H

L

HH

g(x y)

2

2

2

2

2

2


Figure 4 db2 result


(P = 4 R = 10) (P = 8 R = 10) (P = 12 R = 15)

g1

g2

g3

g0gc


(a)

0 20 40 600

2

4

6

8

10

12

14

16

18

20

(b)


119904 (119909) = 1 119909 ge 0

0 119909 lt 0(3)


119875is the



119875




LBPV119875119877

(119870) =

119873

sum

119894=1

119872

sum

119895=1

119882 (LBP119875119877

(119894 119895) 119896) 119896 isin [0 119870]

119882 (LBP119875119877

(119894 119895) 119896) = VAR119875119877

(119894 119895) LBP119875119877

(119894 119895) = 119896

0 otherwise(4)







119894and





119899






119882119894119895

= 119890minus119909119894minus119909119895

2119905

(5)




119883119871119883119879

119886 = 120582119883119863119863119879

119886 (6)


119894119894= sum 119895119882

119895119894 119871 =


0 119886





119909119894997888rarr 119910119894= 119860119879

119909119894 119860 = (119886

0 1198861 119886

119897minus1) (7)





Palm veinimage (I)

sym2(I)

db2(I)

Apx(sym2(I))

Apx(db2(I))

LBPVLBPV(Apx(sym2))

LBPV(I)

LBPV(Apx(db2))

Z-normalization

Z-normalization

Z-normalization

Z-normalization

Z-normalization

Fusio

nFu

sion

LPP WLPP

LPP

Projection

Projection LBPV LPP





119896

119894119894 = 1 sdot sdot sdot 119862



119896is the number



ID119910

= argmin119896

10038171003817100381710038171003817119910 minus 119909119896

119894

10038171003817100381710038171003817

2

(8)









0 02 04 06 08 1 12 14Distance


0123456789

Freq

uenc

e (

)

(a)

FAR ()



10minus4

10minus3

10minus2

10minus1

100

0

05

1

15

2

25

3

FRR

()

(b)


0 02 04 06 08 1 12 140

051

152

253

35

Distance

Freq

uenc

e (

)


(a)

FAR ()


10minus5

10minus4

10minus3

10minus2

10minus1

100

101

102

20

468

1012141618

FRR

()

(b)









0 02 04 06 08 1 12 140

051

152

253

35

454

Distance

Freq

uenc

e (

)


(a)

FAR ()



10minus5

10minus4

10minus3

10minus2

10minus1

100

101

102

02468

101214

FRR

()

(b)


0 02 04 06 08 1 12 140

051

152

253

354

Distance

Freq

uenc

e (

)


(a)

FAR ()



10minus5

10minus4

10minus3

10minus2

10minus1

100

101

102

0102030405060708090

100

FRR

()

(b)











0 02 04 06 08 1 12 140

051

152

253

354

Distance

Freq

uenc

e (

)


(a)

FAR ()



10minus4

10minus3

10minus2

10minus1

100

002040608

112141618

2

FRR

()

(b)













Acknowledgments


References


































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in




Palm veinimage

Palm veinimage

Matchingenhancement

filter

Matchingenhancement

filter

Waveletfeature

extraction


extraction

LBPVfeature

extraction

LBPVfeature

extraction

LPP

LPP LPP

WLPP

WLPP matchingscore

matchingscore

LBPV

Feature extraction

Feature extraction

Fusion


Database

Enrollment steps

Decision

Projection

Projection

Projection

Projection

Verify steps

LBPV

LPPLBPV



119877120590

119865= max (119877

120590

120579119895(119909 119910)) 119895 = 0 1 2 3 4 5

119877120590

120579119895(119909 119910) = 119892

120590

120579119895lowast 119891 (119909 119910)

(2)







LBP119875119877

=

119875minus1

sum

119901=0

119904 (119892119875

minus 119892119888) 2119875


Original image

048 05205 054 056 058 06206 064 0660

20

40

60


(a)

Enhancement image

01 02 03 04 05 06 07 08 090 10

20

40

60

80

100


(b)


Image

LL

LL

LH

HL

HH


H

L

HH

g(x y)

2

2

2

2

2

2


Figure 4 db2 result


(P = 4 R = 10) (P = 8 R = 10) (P = 12 R = 15)

g1

g2

g3

g0gc


(a)

0 20 40 600

2

4

6

8

10

12

14

16

18

20

(b)


119904 (119909) = 1 119909 ge 0

0 119909 lt 0(3)


119875is the



119875




LBPV119875119877

(119870) =

119873

sum

119894=1

119872

sum

119895=1

119882 (LBP119875119877

(119894 119895) 119896) 119896 isin [0 119870]

119882 (LBP119875119877

(119894 119895) 119896) = VAR119875119877

(119894 119895) LBP119875119877

(119894 119895) = 119896

0 otherwise(4)







119894and





119899






119882119894119895

= 119890minus119909119894minus119909119895

2119905

(5)




119883119871119883119879

119886 = 120582119883119863119863119879

119886 (6)


119894119894= sum 119895119882

119895119894 119871 =


0 119886





119909119894997888rarr 119910119894= 119860119879

119909119894 119860 = (119886

0 1198861 119886

119897minus1) (7)





Palm veinimage (I)

sym2(I)

db2(I)

Apx(sym2(I))

Apx(db2(I))

LBPVLBPV(Apx(sym2))

LBPV(I)

LBPV(Apx(db2))

Z-normalization

Z-normalization

Z-normalization

Z-normalization

Z-normalization

Fusio

nFu

sion

LPP WLPP

LPP

Projection

Projection LBPV LPP





119896

119894119894 = 1 sdot sdot sdot 119862



119896is the number



ID119910

= argmin119896

10038171003817100381710038171003817119910 minus 119909119896

119894

10038171003817100381710038171003817

2

(8)









0 02 04 06 08 1 12 14Distance


0123456789

Freq

uenc

e (

)

(a)

FAR ()



10minus4

10minus3

10minus2

10minus1

100

0

05

1

15

2

25

3

FRR

()

(b)


0 02 04 06 08 1 12 140

051

152

253

35

Distance

Freq

uenc

e (

)


(a)

FAR ()


10minus5

10minus4

10minus3

10minus2

10minus1

100

101

102

20

468

1012141618

FRR

()

(b)









0 02 04 06 08 1 12 140

051

152

253

35

454

Distance

Freq

uenc

e (

)


(a)

FAR ()



10minus5

10minus4

10minus3

10minus2

10minus1

100

101

102

02468

101214

FRR

()

(b)


0 02 04 06 08 1 12 140

051

152

253

354

Distance

Freq

uenc

e (

)


(a)

FAR ()



10minus5

10minus4

10minus3

10minus2

10minus1

100

101

102

0102030405060708090

100

FRR

()

(b)











0 02 04 06 08 1 12 140

051

152

253

354

Distance

Freq

uenc

e (

)


(a)

FAR ()



10minus4

10minus3

10minus2

10minus1

100

002040608

112141618

2

FRR

()

(b)













Acknowledgments


References


































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in




Original image

048 05205 054 056 058 06206 064 0660

20

40

60


(a)

Enhancement image

01 02 03 04 05 06 07 08 090 10

20

40

60

80

100


(b)


Image

LL

LL

LH

HL

HH


H

L

HH

g(x y)

2

2

2

2

2

2


Figure 4 db2 result


(P = 4 R = 10) (P = 8 R = 10) (P = 12 R = 15)

g1

g2

g3

g0gc


(a)

0 20 40 600

2

4

6

8

10

12

14

16

18

20

(b)


119904 (119909) = 1 119909 ge 0

0 119909 lt 0(3)


119875is the



119875




LBPV119875119877

(119870) =

119873

sum

119894=1

119872

sum

119895=1

119882 (LBP119875119877

(119894 119895) 119896) 119896 isin [0 119870]

119882 (LBP119875119877

(119894 119895) 119896) = VAR119875119877

(119894 119895) LBP119875119877

(119894 119895) = 119896

0 otherwise(4)







119894and





119899






119882119894119895

= 119890minus119909119894minus119909119895

2119905

(5)




119883119871119883119879

119886 = 120582119883119863119863119879

119886 (6)


119894119894= sum 119895119882

119895119894 119871 =


0 119886





119909119894997888rarr 119910119894= 119860119879

119909119894 119860 = (119886

0 1198861 119886

119897minus1) (7)





Palm veinimage (I)

sym2(I)

db2(I)

Apx(sym2(I))

Apx(db2(I))

LBPVLBPV(Apx(sym2))

LBPV(I)

LBPV(Apx(db2))

Z-normalization

Z-normalization

Z-normalization

Z-normalization

Z-normalization

Fusio

nFu

sion

LPP WLPP

LPP

Projection

Projection LBPV LPP





119896

119894119894 = 1 sdot sdot sdot 119862



119896is the number



ID119910

= argmin119896

10038171003817100381710038171003817119910 minus 119909119896

119894

10038171003817100381710038171003817

2

(8)









0 02 04 06 08 1 12 14Distance


0123456789

Freq

uenc

e (

)

(a)

FAR ()



10minus4

10minus3

10minus2

10minus1

100

0

05

1

15

2

25

3

FRR

()

(b)


0 02 04 06 08 1 12 140

051

152

253

35

Distance

Freq

uenc

e (

)


(a)

FAR ()


10minus5

10minus4

10minus3

10minus2

10minus1

100

101

102

20

468

1012141618

FRR

()

(b)









0 02 04 06 08 1 12 140

051

152

253

35

454

Distance

Freq

uenc

e (

)


(a)

FAR ()



10minus5

10minus4

10minus3

10minus2

10minus1

100

101

102

02468

101214

FRR

()

(b)


0 02 04 06 08 1 12 140

051

152

253

354

Distance

Freq

uenc

e (

)


(a)

FAR ()



10minus5

10minus4

10minus3

10minus2

10minus1

100

101

102

0102030405060708090

100

FRR

()

(b)











0 02 04 06 08 1 12 140

051

152

253

354

Distance

Freq

uenc

e (

)


(a)

FAR ()



10minus4

10minus3

10minus2

10minus1

100

002040608

112141618

2

FRR

()

(b)













Acknowledgments


References


































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in




(P = 4 R = 10) (P = 8 R = 10) (P = 12 R = 15)

g1

g2

g3

g0gc


(a)

0 20 40 600

2

4

6

8

10

12

14

16

18

20

(b)


119904 (119909) = 1 119909 ge 0

0 119909 lt 0(3)


119875is the



119875




LBPV119875119877

(119870) =

119873

sum

119894=1

119872

sum

119895=1

119882 (LBP119875119877

(119894 119895) 119896) 119896 isin [0 119870]

119882 (LBP119875119877

(119894 119895) 119896) = VAR119875119877

(119894 119895) LBP119875119877

(119894 119895) = 119896

0 otherwise(4)







119894and





119899






119882119894119895

= 119890minus119909119894minus119909119895

2119905

(5)




119883119871119883119879

119886 = 120582119883119863119863119879

119886 (6)


119894119894= sum 119895119882

119895119894 119871 =


0 119886





119909119894997888rarr 119910119894= 119860119879

119909119894 119860 = (119886

0 1198861 119886

119897minus1) (7)





Palm veinimage (I)

sym2(I)

db2(I)

Apx(sym2(I))

Apx(db2(I))

LBPVLBPV(Apx(sym2))

LBPV(I)

LBPV(Apx(db2))

Z-normalization

Z-normalization

Z-normalization

Z-normalization

Z-normalization

Fusio

nFu

sion

LPP WLPP

LPP

Projection

Projection LBPV LPP





119896

119894119894 = 1 sdot sdot sdot 119862



119896is the number



ID119910

= argmin119896

10038171003817100381710038171003817119910 minus 119909119896

119894

10038171003817100381710038171003817

2

(8)









0 02 04 06 08 1 12 14Distance


0123456789

Freq

uenc

e (

)

(a)

FAR ()



10minus4

10minus3

10minus2

10minus1

100

0

05

1

15

2

25

3

FRR

()

(b)


0 02 04 06 08 1 12 140

051

152

253

35

Distance

Freq

uenc

e (

)


(a)

FAR ()


10minus5

10minus4

10minus3

10minus2

10minus1

100

101

102

20

468

1012141618

FRR

()

(b)









0 02 04 06 08 1 12 140

051

152

253

35

454

Distance

Freq

uenc

e (

)


(a)

FAR ()



10minus5

10minus4

10minus3

10minus2

10minus1

100

101

102

02468

101214

FRR

()

(b)


0 02 04 06 08 1 12 140

051

152

253

354

Distance

Freq

uenc

e (

)


(a)

FAR ()



10minus5

10minus4

10minus3

10minus2

10minus1

100

101

102

0102030405060708090

100

FRR

()

(b)











0 02 04 06 08 1 12 140

051

152

253

354

Distance

Freq

uenc

e (

)


(a)

FAR ()



10minus4

10minus3

10minus2

10minus1

100

002040608

112141618

2

FRR

()

(b)













Acknowledgments


References


































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in







119894and





119899






119882119894119895

= 119890minus119909119894minus119909119895

2119905

(5)




119883119871119883119879

119886 = 120582119883119863119863119879

119886 (6)


119894119894= sum 119895119882

119895119894 119871 =


0 119886





119909119894997888rarr 119910119894= 119860119879

119909119894 119860 = (119886

0 1198861 119886

119897minus1) (7)





Palm veinimage (I)

sym2(I)

db2(I)

Apx(sym2(I))

Apx(db2(I))

LBPVLBPV(Apx(sym2))

LBPV(I)

LBPV(Apx(db2))

Z-normalization

Z-normalization

Z-normalization

Z-normalization

Z-normalization

Fusio

nFu

sion

LPP WLPP

LPP

Projection

Projection LBPV LPP





119896

119894119894 = 1 sdot sdot sdot 119862



119896is the number



ID119910

= argmin119896

10038171003817100381710038171003817119910 minus 119909119896

119894

10038171003817100381710038171003817

2

(8)









0 02 04 06 08 1 12 14Distance


0123456789

Freq

uenc

e (

)

(a)

FAR ()



10minus4

10minus3

10minus2

10minus1

100

0

05

1

15

2

25

3

FRR

()

(b)


0 02 04 06 08 1 12 140

051

152

253

35

Distance

Freq

uenc

e (

)


(a)

FAR ()


10minus5

10minus4

10minus3

10minus2

10minus1

100

101

102

20

468

1012141618

FRR

()

(b)









0 02 04 06 08 1 12 140

051

152

253

35

454

Distance

Freq

uenc

e (

)


(a)

FAR ()



10minus5

10minus4

10minus3

10minus2

10minus1

100

101

102

02468

101214

FRR

()

(b)


0 02 04 06 08 1 12 140

051

152

253

354

Distance

Freq

uenc

e (

)


(a)

FAR ()



10minus5

10minus4

10minus3

10minus2

10minus1

100

101

102

0102030405060708090

100

FRR

()

(b)











0 02 04 06 08 1 12 140

051

152

253

354

Distance

Freq

uenc

e (

)


(a)

FAR ()



10minus4

10minus3

10minus2

10minus1

100

002040608

112141618

2

FRR

()

(b)













Acknowledgments


References


































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in




Palm veinimage (I)

sym2(I)

db2(I)

Apx(sym2(I))

Apx(db2(I))

LBPVLBPV(Apx(sym2))

LBPV(I)

LBPV(Apx(db2))

Z-normalization

Z-normalization

Z-normalization

Z-normalization

Z-normalization

Fusio

nFu

sion

LPP WLPP

LPP

Projection

Projection LBPV LPP





119896

119894119894 = 1 sdot sdot sdot 119862



119896is the number



ID119910

= argmin119896

10038171003817100381710038171003817119910 minus 119909119896

119894

10038171003817100381710038171003817

2

(8)









0 02 04 06 08 1 12 14Distance


0123456789

Freq

uenc

e (

)

(a)

FAR ()



10minus4

10minus3

10minus2

10minus1

100

0

05

1

15

2

25

3

FRR

()

(b)


0 02 04 06 08 1 12 140

051

152

253

35

Distance

Freq

uenc

e (

)


(a)

FAR ()


10minus5

10minus4

10minus3

10minus2

10minus1

100

101

102

20

468

1012141618

FRR

()

(b)









0 02 04 06 08 1 12 140

051

152

253

35

454

Distance

Freq

uenc

e (

)


(a)

FAR ()



10minus5

10minus4

10minus3

10minus2

10minus1

100

101

102

02468

101214

FRR

()

(b)


0 02 04 06 08 1 12 140

051

152

253

354

Distance

Freq

uenc

e (

)


(a)

FAR ()



10minus5

10minus4

10minus3

10minus2

10minus1

100

101

102

0102030405060708090

100

FRR

()

(b)











0 02 04 06 08 1 12 140

051

152

253

354

Distance

Freq

uenc

e (

)


(a)

FAR ()



10minus4

10minus3

10minus2

10minus1

100

002040608

112141618

2

FRR

()

(b)













Acknowledgments


References


































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in




0 02 04 06 08 1 12 14Distance


0123456789

Freq

uenc

e (

)

(a)

FAR ()



10minus4

10minus3

10minus2

10minus1

100

0

05

1

15

2

25

3

FRR

()

(b)


0 02 04 06 08 1 12 140

051

152

253

35

Distance

Freq

uenc

e (

)


(a)

FAR ()


10minus5

10minus4

10minus3

10minus2

10minus1

100

101

102

20

468

1012141618

FRR

()

(b)









0 02 04 06 08 1 12 140

051

152

253

35

454

Distance

Freq

uenc

e (

)


(a)

FAR ()



10minus5

10minus4

10minus3

10minus2

10minus1

100

101

102

02468

101214

FRR

()

(b)


0 02 04 06 08 1 12 140

051

152

253

354

Distance

Freq

uenc

e (

)


(a)

FAR ()



10minus5

10minus4

10minus3

10minus2

10minus1

100

101

102

0102030405060708090

100

FRR

()

(b)











0 02 04 06 08 1 12 140

051

152

253

354

Distance

Freq

uenc

e (

)


(a)

FAR ()



10minus4

10minus3

10minus2

10minus1

100

002040608

112141618

2

FRR

()

(b)













Acknowledgments


References


































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in




0 02 04 06 08 1 12 140

051

152

253

35

454

Distance

Freq

uenc

e (

)


(a)

FAR ()



10minus5

10minus4

10minus3

10minus2

10minus1

100

101

102

02468

101214

FRR

()

(b)


0 02 04 06 08 1 12 140

051

152

253

354

Distance

Freq

uenc

e (

)


(a)

FAR ()



10minus5

10minus4

10minus3

10minus2

10minus1

100

101

102

0102030405060708090

100

FRR

()

(b)











0 02 04 06 08 1 12 140

051

152

253

354

Distance

Freq

uenc

e (

)


(a)

FAR ()



10minus4

10minus3

10minus2

10minus1

100

002040608

112141618

2

FRR

()

(b)













Acknowledgments


References


































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in




0 02 04 06 08 1 12 140

051

152

253

354

Distance

Freq

uenc

e (

)


(a)

FAR ()



10minus4

10minus3

10minus2

10minus1

100

002040608

112141618

2

FRR

()

(b)













Acknowledgments


References


































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in

































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in



Research Article Palm Vein Verification Using Multiple...

Documents

Transcript of Research Article Palm Vein Verification Using Multiple...