research article performance evaluation of multimodal...

Research ArticlePerformance Evaluation of MultimodalMultifeature Authentication System Using KNN Classification

Gayathri Rajagopal1 and Ramamoorthy Palaniswamy2

1Department of Electronics and Communication Engineering Sri Venkateswara College of Engineering Anna UniversitySriperumbudur 602117 India2Department of Electronics and Communication Engineering Aditya Institute of Technology Coimbatore 641107 India

Correspondence should be addressed to Gayathri Rajagopal gayathricontactgmailcom

Received 17 August 2015 Revised 19 October 2015 Accepted 25 October 2015

Academic Editor Michele Nappi

Copyright copy 2015 G Rajagopal and R Palaniswamy This is an open access article distributed under the Creative CommonsAttribution License which permits unrestricted use distribution and reproduction in any medium provided the original work isproperly cited

This research proposes a multimodal multifeature biometric system for human recognition using two traits that is palmprint andirisThe purpose of this research is to analyse integration ofmultimodal andmultifeature biometric systemusing feature level fusionto achieve better performance The main aim of the proposed system is to increase the recognition accuracy using feature levelfusion The features at the feature level fusion are raw biometric data which contains rich information when compared to decisionand matching score level fusion Hence information fused at the feature level is expected to obtain improved recognition accuracyHowever information fused at feature level has the problem of curse in dimensionality here PCA (principal component analysis)is used to diminish the dimensionality of the feature sets as they are high dimensional The proposed multimodal results werecompared with other multimodal and monomodal approaches Out of these comparisons the multimodal multifeature palmprintiris fusion offers significant improvements in the accuracy of the suggested multimodal biometric systemThe proposed algorithmis tested using created virtual multimodal database using UPOL iris database and PolyU palmprint database

1 Introduction

Amultimodal biometric system fuses the evidences presentedby multiple biometric traits Multimodal biometric tech-niques have received the best recognition because additionalinformation between different biometrics could get improvedaccuracy To obtain a successful multibiometric system onehas to essentially implement a good fusingmethodology suchas match score feature and decision level fusion

In order to reduce the error rate and to improve the per-formance accuracy many researchers worked onmultimodalbiometric system Hariprasath and Prabakar [1] proposeda multimodal biometric system using iris and palmprintbased on score level fusion and authentication is obtainedby hamming distance method Gargouri Ben Ayed et al [2]fused fingerprint and faces using match score level fusionusingweighted summethodHereGaborwavelet network forface and LBP fingerprint features are fused Abdolahi et al [3]proposed fuzzy basedmultimodal biometric system by fusing

iris and fingerprint using decision level fusion to provideimproved recognition rate Bahgat et al [4] fused palm veinand face biometric to obtain the better recognition rate

Various multimodal score level fusion schemes wereproposed by different researchers Baig et al [5] proposedscore level fusion of iris and fingerprint which is classifiedusing hamming distance calculationWang et al [6] proposeda score level based multimodal biometric combining iris andpalmprint using Gaussian mixture model Vatsa et al [7]combined multi-instant and multiunit iris verification Wangand Han [8] fused iris and face using score level fusion inwhich different scores are obtained for different traits and theobtained scores are combined using Support VectorMachineWang and Han [9] Kayaoglu et al [10] Zhang et al [11]and Peng et al [12] investigated multimodal biometric fusionusing decision and score level fusion

Monwar and Gavrilova [13] investigated rank level fusionof face ear and signature using principal component anal-ysis and Fisherrsquos linear discriminant methods Kumar and

Hindawi Publishing Corporatione Scientific World JournalVolume 2015 Article ID 762341 9 pageshttpdxdoiorg1011552015762341

2 The Scientific World Journal

Shekhar [14] investigated multiple palmprint recognitionusing rank level fusion which uses borda count bucklinhighest rank and logistic regressionMatch score level fusionusing feed forward neural network for the fusion of face andpalmprint has been investigated by Thepade and Bhondave[15] Yang et al [16] investigated multiple dependency ofpalmprint using feature level fusion and score level fusionWang and Han [9] investigated face iris fusion using scorelevel fusion Jain et al [17] investigated the multimodalbiometric system based on the face and hand geometrybiometrics at the score level

Conti et al [18] proposed a multimodal biometric basedon two-fingerprint acquisition which uses score level fusionand obtained an improvement of 6 when compared tomonomodal biometric based system Yang et al [19] usedmatched score level fusion to fuse hand geometry fingerprintand palmprint multimodal biometric system He used aself-constructed database of 97 subjects Besbes et al [20]proposed a hybrid multitrait biometric method using irisand fingerprint Yang et al [16] proposed decision levelfusion fingerprint templates Here assessment was taken byindividual unimodal assessment through an ldquoANDrdquo operator

Most significant contribution published in recent yearspertaining to multimodal biometric fusion focused exten-sively on fusing data at the matching score level and decisionlevel It has been observed thatmost of the important featuresare lost on performing data fusion at the latter stages (matchscore level and decision level) In spite of the abundanceof investigations related to multimodal biometrics relativelylittle work was done at feature level fusion since featurefusion has rich information content compared to fusionat the later stages Therefore the current exploration on amultimodal biometric fusion at the feature level is anticipatedto attain improved recognition accuracy compared to thefusion at the later stages

2 Feature Fusion Using HierarchicalMultiresolution LBP and Gabor

This research mainly discusses the multifeature fusion ofpalmprint and iris biometrics using feature level fusion HereFigure 1 illustrates feature fusion using hierarchical multires-olution LBP and Gabor It consists of three major blockspreprocessing feature extraction and fusion Multimodalmultifeature-based biometric system involves the followingsteps

(i) The two modalities presumed are palmprint and irisimage which are given as input

(ii) The Gabor feature and hierarchical multiresolutionLBP features of palmprint and iris image respectivelyare taken

(iii) Images are fused by using feature level fusion(iv) 119870 nearest neighbor is used for classification(v) Recognition accuracy is calculated

21 Gabor Wavelets Gabor wavelets are a filter bank con-sisting of Gabor filters with diverse scales and rotation It is

Registration

Identificationverification

Biometric modality 1

Preprocessing and ROI extraction



Feature level fusion

Multifeature extraction

Matching

Database


Figure 1 Feature fusion using hierarchical multiresolution LBP andGabor

efficient for analyzing dissimilar phased features like abruptridges or edges Gabor space is extremely useful in variousmedical image-processing purposes (Lades et al [21])Mainlythe Gabor wavelets were developed to represent the receptivefields of simple cells in the visual cortex However in practicethey confine to most of the salient properties togetherwith frequency selectivity orientation selectivity and spatiallocalization Actually here the image is conlvolved with abank of Gabor filters of different orientations and scalesGabor wavelet has the following general form as in

1205951205831]1(119911)

=

100381710038171003817100381710038171198701205831]1

10038171003817100381710038171003817

2

1205902119890minus1198701205831]1 2119885221205902

[1198901198951198701205831]1119885 minus 119890

minus12059022]

(1)

where represents the norm operator ]1 and 1205831 are scaleand the orientation respectively of the Gabor kernel 119911 =

(1199091 1199101) represents a variable in spatial domain and 1198701205831]1

represents wave vector and 120590 is the standard deviationThe wave vector is represented in

1198701205831]1

= 119870]1

(cos1206011205831

+ 119895 sin1206011205831

) (2)

where 119870]1

= 119870max119891]1

and 1206011205831

= 1205831205878 with 119870max being themaximum frequency and 119891 is the spacing factor In thisresearch the Gabor kernel filter used is of three differentscales and four orientations Figure 2 shows the Gabor kernelfilter output

22 Hierarchical Multiresolution Local Binary Pattern Ojalaet al [22] introduced local binary pattern in 1996 The localbinary pattern is a gray scale invariant texture measureand is a helpful tool to model texture images It tags thepixels representation by using threshold of the pixels ofthe local neighbor around each pixel and considers theresult as binary numbers It is a combining approach todivergent statistical and structural forms of texture anal-ysis The major property of hierarchical multiresolutionLBP is its robustness to monotonic gray scale alterationscaused An added advantage of hierarchical multiresolutionLBP is its computational simplicity to analyze images in

The Scientific World Journal 3

Figure 2 Gabor Kernal Filer Output

real time Hierarchical multiresolution LBP is operated witheight neighbors of a pixel with the value of the middlepixel as a threshold Hierarchical multiresolution LBP codesfor a neighbor are produced by multiplying the thresholdassessment with weights specified to the resultant pixels andthe results are summated It is executed by an orthogonalmeasure of local contrast The averages of gray levels underthe middle pixel are deduced from that of the gray levels overthe center pixel Two-dimensional distributions hierarchicalmultiresolution LBP and local contrast measures are used asfeatures

Local binary pattern (Ojala et al [22]) is used to capturethe local structure of the image Center pixel of the image isassumed to be 119870119888 = (119909 119910) with 8 neighboring pixels (119872 =

8) and radius of the neighborhood is assumed as 119903 = 1 Thehierarchical multiresolution LBP is obtained as given in

LBP119872119903119870119888 =119872minus1

sum

119898

119891 (119870119899 119870119888) 2119899

119891 (119870119899 119870119888) =

1

0

119868 (119870119899) ge 119868 (119870119888)

119868 (119870119899) gt 119868 (119870119888)

(3)

where 119868(119870119899) and 119868(119870119899) are the gray values of the center pixel119870119888 = (119909 119910)

Gray values of119872 neighbouring pixels are obtained usingbilinear interpolation and the coordinate of119870119899 is determinedby

(119909119899 119910119899) = [119909119888 + 119903 cos(2120587119899119872

) 119910119888 minus 119903 sin(2120587119899119872

)] (4)

To enhance the performance of LBP operator mul-tiresolution LBP features are used Multiresolution LBPfeatures consist of richer information than the single LBPoperator Conventionally LBP features with different scalesare obtained and concatenated into a lengthy feature Theobtained feature contains enormous information but it has adrawback of curse of dimensionality

Nonuniform pattern contains more useful informationsome of the processing steps are investigated by Raja andGong [23] and Liao et al [24] However the recognition

Figure 3 Binary pattern of different radius Filled pattern represents1 while the blank circle represents 0

accuracy depends on the training samples Figure 3 explainsan illustration of the binary pattern It consists of nonuniform(bigger radius) and uniform (smaller radius) patterns Forthe uniform pattern a subhistogram is constructed but forthe nonuniform pattern they are processed to dig out theirLBP pattern by smaller dimension Thus the processingsteps are continued until the pixels patterns are uniformFigure 4 explains the proposed multiresolution hierarchicalLBP system Initially LBP histogram is constructed Usingnonuniform pattern for 119877 = 3 a new histogram patternof 119877 = 2 is constructed Then using nonuniform pattern119877 = 2 are processed to obtain the histogram pattern of119877 = 1

3 Proposed Multimodal Feature FusionBlock Diagram

The proposed methodology for investigating the multimodalmultifeature biometric systems is based on the combinationof palmprint and iris Feature fusion has the advantage ofexploiting rich information from each biometric Figure 5represents proposed feature fusion multimodal biometricsystem based on Gabor and hierarchical multiresolutionLBP extraction The feature vectors are extracted indepen-dently from the preprocessed images of palmprint and irisThese features are normalized to obtain a single vectorThe feature vectors of input images (test image) are thenevaluated with the templates of the database (train image)to produce the output Fusing more than one modalityimproves the recognition accuracy reduces False AcceptanceRate and False Rejection Rate The proposed multimodalmultifeature biometric method overcomes the restrictionsof single biometric systems and convenes the accuracyrequirements

Figure 6 explains the original image of the iris and palm-print taken fromUPOL and PolyU palmprint database Herevarious stages of palmprint and iris image processing areexplained that is preprocessing of palmprint and iris image


Modified multiresolution scheme

Furtherprocessing

Furtherprocessing

Nonuniform pattern of


R = 3

R = 2

Uniform pattern of R = 3



R = 1

R = 2

R = 3 middot middot middot

middot middot middot


Figure 4 An illustration of proposed hierarchical multiresolution system

Iris input

Feature extraction using Gabor and LBP




Palmprint input Iris inputPalmprint input




Preprocessing and ROIextraction

Normalize the feature vector to single vector

Normalize the feature vector to get single vector

Normalize the feature vector to

single vector


single vector

Matching Decision

Training phase Testing phase

KNN classificationFeature fusion

Database

Feature fusion

Figure 5 Proposed feature fusion multimodal biometric system

feature level fused image and segmentation result of thefused image The proposed multimodal biometric techniqueexploits most of the information from each monomodalbiometric Gabor and hierarchical multiresolution LBP fea-tures are extracted for each palmprint and iris image andthe acquired features are fused by using feature fusion andstored in a database for matching Figure 7 illustrates thephase congruency and gradient magnitude extracted fromtest image and the matched image is stored in a databaseFigure 8 illustrates a sample image found in the databaseduring matching

4 Result and Discussion

To evaluate the effectiveness of the proposed multimodalbiometric system a database containing palmprint andiris samples are required To build the virtual multimodaldatabase images are adopted fromPolyUPalmprint databaseIt includes 7752 images corresponding to 386 subjects Irisimage databases are adopted fromUPOLdatabase It includes768 images of 576 times 768 pixels captured from 128 subjects intwo distinct sessions Later each sample of the iris database israndomlymergedwith one sample of the palmprint database


(a) (b)

(c) (d)

(e) (f)

Figure 6 Feature fused image (a) eye image (b) preprocessed eye image (c) palmprint image (d) preprocessed palmprint image (e) featurefused image and (f) segmented image

For the research work 123 individual palmprint imagesand iris images are selected every person has 5 samplesand totaling up to 615 Each personrsquos palmprint and irisimages were taken as a template (totaling 123)The remaining492 were used as training samples The experiments wereperformed in MATLAB with image processing Toolboxon a device with an Intel core 2 Duo CPU processorHere among 123 dissimilar test database untrained imagesexperience similar algorithm as trained image and compareto the original trained image Figure 9 explains the119870 nearestneighbor classification result of the proposed multimodalbiometric fusion of palmprint and iris Here legends with ldquo119900rdquo

of different colours represent the test data of 123 individualsSymbol ldquolowastrdquo represents 492 trained samples of 123 individuals119870NN classification is obtained based on the multifeaturefusion (Gabor and hierarchical multiresolution LBP) valueof the test and trained image The proposed multifeaturefusion method based on hierarchical multiresolution LBPand Gabor fusing iris and palmprint system achieves arecognition accuracy of 9998 with equal error rate (ERR)of 00378

Twenty samples were taken and analyzed using 119870meansalgorithm Sixteen samples were analyzed using 119870 nearestneighbor classification algorithm Table 1 represents two class


(a) (b)

(c) (d)

Figure 7 Phase congruency of (a) test image and (b) matched image Gradient magnitude of (c) test image and (d) matched image

Figure 8 Sample image found during matching

ids assumed for 119870 means algorithm Here each class idwas assumed to have ten classes Table 2 represents the

Table 1 Assumed class id for 119870-means algorithm

3 2 4 4 3 1 4 4 3 3 Class id 11 1 2 2 3 1 5 4 5 5 Class id 2

matching accuracy obtained for each sample using119870meansclassification algorithm Here 20 samples S1 to S20 wereconsidered Table 3 represents the class id assumed for 119870

nearest neighborhood classification Here each class id wasassumed to have four classes Table 4 represents thematchingscores obtained by using119870 nearest neighborh algorithm

Here class id one and class id four werematched perfectlybecause they both belong to same class It was found that119870-nearest neighbor algorithm obtained a higher matchingaccuracy than the119870means algorithm

Figure 10 shows the receiver operating charachteristics(ROC) curve for the unimodal and bimodal biometric sys-tem From the graph it has been observed that the proposedmultimodal biometric system acheives a reduced equal errorrate (EER) of 00378


KNN classification based on feature fusion of palmprint and iris

123456

789101112

131415161718

1920212223

0095

0096

0097

0098

0099

01

0101

0102

0103

0104

0105

Scor

e

0104 0106 0108 011 0112 0114 0116 01180102Data

Figure 9 119870NN classification for the proposed multifeature fusionmultimodal biometric

Table 2 Matching scores using119870-means algorithm

S1 S2 S3 S4 S5 S6 S7 S8 S9 S10749 7281 7699 7221 0 0 0 0 0 0789 7351 7239 7188 0 0 0 0 0 00 0 0 0 8842 8578 7343 8771 0 00 0 0 0 8776 8611 6684 8897 0 00 0 0 0 0 0 0 0 7163 7234742 6921 7392 7782 0 0 0 0 0 00 0 0 0 0 0 0 0 0 00 0 0 0 0 0 0 0 0 00 0 0 0 0 0 0 0 0 00 0 0 0 0 0 0 0 0 0S11 S12 S13 S14 S15 S16 S17 S18 S19 S200 0 0 0 0 0 0 0 0 00 0 0 0 0 0 0 0 0 00 0 0 0 0 0 0 0 0 00 0 0 0 0 0 0 0 0 07546 7418 0 0 0 0 0 0 0 00 0 0 0 0 0 0 0 0 00 0 0 0 0 0 7379 8644 7002 76720 0 6920 7620 8865 8629 0 0 0 00 0 0 0 0 0 6771 8885 7372 76090 0 0 0 0 0 6819 8523 6902 6976

Table 5 explains the comparison of different modalitycombinations and their recognition accuracy From theclassified result it was concluded that the performance of

Table 3 Assumed class id for 119870-nearest neighbor algorithm

1 1 1 4 Class id 11 2 3 4 Class id 2

Table 4 Matching scores using 119870-nearest neighbor algorithm

S1 S2 S3 S4 S5 S6 S7 S89918 8234 8130 9845 0 0 0 00 0 0 0 9843 8876 8683 99260 0 0 0 0 0 0 00 0 0 0 0 0 0 0S9 S10 S11 S12 S13 S14 S15 S160 0 0 0 0 0 0 00 0 0 0 0 0 0 09634 8543 8334 9843 0 0 0 00 0 0 0 9334 8543 8646 9489

1 iris + palmprint2 iris3 palmprint

0002004006008

01Fa

lse re

ject

ion

rate

002 004 006 008 010False acceptance rate

EER = 00528

EER = 0042

EER = 00378

Figure 10 ROC curves for the unimodal and multimodal system

the proposed iris palmprint features fusion obtains bet-ter recognition accuracy when compared to other fusionmethods Here feature fusion offers enhanced performancecompared to other level of fusion Moreover multifeature(hierarchical multiresolution LBP and Gabor) multimodal(palmprint and iris) feature fusion increases the recognitionaccuracyThe combination of palmprint and iris (multimodalmultifeature fusion) is classified using 119870 nearest neighborhere the distance between test and trained vectors is smallwhen compared to the other combinations discussed so far

5 Conclusion

This research has presented a feature level fusion of multi-modal multifeature palmprint and iris recognition systemGabor wavelets and hierarchical multiresolution LBP areused for feature extraction and PCA was applied to reducethe dimensionality Finally the feature vectors are classifiedusing 119870NN The experiment result of the proposed multi-feature fusion method based on multiresolution hierarchicalmultiresolution LBP and Gabor fusing iris and palmprintsystem achieves a recognition accuracy of 9996 withequal error rate of 00378 on the publicly available PolyU


Table 5 Comparison of various modalities

Method Recognition accuracy ModalitiesFeature fusion of single scale LBPGuo et al [25] 8146 Face and palmprint

Score level fusionZhou and Bhanu [26] 9330 Side face and gait

Feature fusion of multiresolution LBPGuo et al [25] 9479 Face and palmprint

Score level fusionKumar et al [27] 9459 Hand geometry and palmprint

Score level fusionNandakumar et al [28] 9480 Fingerprint and iris

Feature Fusion of modified multiresolutionGuo et al [25] 9667 Face and palmprint

Score level fusionZhang et al [29] 9267 Fingerprint and palmprint

Score level fusionKorves et al [30] 9750 Fingerprint and face

Decision level fusionAbdolahi et al [3] 9820 Fingerprint and iris

Feature fusionZhou and Bhanu [26] 9740 Side face and gait

Score level fusionAguilar et al [31] 9820 Iris and palmprint

Rank level fusionMonwar and Gavrilova [13] 9882 Face ear and signature

Proposed feature fusion of hierarchicalmultiresolution LBP and Gabor 9996 Iris and palmprint

palmprint and UPOL iris database Here feature fusionoffers enhanced performance compared to other levels offusion Moreover multifeature (hierarchical multiresolutionLBP and Gabor) multimodal (palmprint and Iris) featurefusion increases the recognition accuracy The combinationof palmprint and iris (multimodal multifeature fusion) isclassified using119870nearest neighbor here the distance betweentest and trained vectors is small when compared to othercombinations discussed so far

Conflict of Interests

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

References

[1] S Hariprasath and T N Prabakar ldquoMultimodal biometric rec-ognition using iris feature extraction and palmprint featuresrdquoin Proceedings of the 1st International Conference on Advances inEngineering Science and Management (ICAESM rsquo12) pp 174ndash179 March 2012

[2] N Gargouri Ben Ayed A D Masmoudi and D S MasmoudildquoA new human identification based on fusion fingerprints andfaces biometrics using LBP and GWN descriptorsrdquo in Proceed-ings of the 8th InternationalMulti-Conference on Systems Signalsand Devices (SSD rsquo11) pp 1ndash7 Sousse Tunisia March 2011

[3] M Abdolahi M Mohamadi and M Jafari ldquoMultimodal bio-metric system fusion using fingerprint and iris with fuzzy logicrdquo

International Journal of Soft Computing and Engineering vol 2no 6 pp 504ndash510 2013

[4] S F Bahgat S Ghoniemy and M Alotaibi ldquoProposed multi-modal palm veins-face biometric authenticationrdquo InternationalJournal of Advanced Computer Science and Applications vol 4no 6 2013

[5] A Baig A Bouridane F Kurugollu and G Qu ldquoFingerprint-iris fusion based identification system using a single Hammingdistancerdquo in Proceedings of the International Symposium on Bio-inspired Learning and Intelligent Systems for Security (BLISSrsquo09) pp 9ndash12 Edinburgh UK August 2009

[6] J Wang Y Li X Ao C Wang and J Zhou ldquoMulti-modal bi-ometric authentication fusing iris and palmprint based onGMMrdquo in Proceedings of the 15th IEEESP Workshop on Sta-tistical Signal Processing (SSP rsquo09) pp 349ndash352 Cardiff WalesAugust-September 2009

[7] M Vatsa R Singh A Noore and S K Singh ldquoBelief functiontheory based biometric match score fusion case studies inmulti-instance and multi-unit iris verificationrdquo in Proceedingsof the 7th International Conference on Advances in PatternRecognition (ICAPR rsquo09) pp 433ndash436 Kolkata India February2009

[8] F Wang and J Han ldquoMultimodal biometric authenticationbased on score level fusion using support vector machinerdquoOpto-Electronics Review vol 17 no 1 pp 59ndash64 2009

[9] F Wang and J Han ldquoRobust multimodal biometric authentica-tion integrating iris face and palmprintrdquo Information Technol-ogy and Control vol 37 no 4 2015


[10] M Kayaoglu B Topcu and U Uludag ldquoBiometric matchingand fusion system for fingerprints from non-distal phalangesrdquohttparxivorgabs150504028

[11] D Zhang F Song Y Xu and Z LiangAdvanced Pattern Recog-nition Technologies with Applications to Biometrics MedicalInformation Science Reference IGI Global 2009

[12] J Peng Q Li Q Han and X Niu ldquoA new approach for fingermultimodal biometric verification based on score-level fusionrdquoIEICE Transactions on Information and Systems vol E96-D no8 pp 846ndash859 2013

[13] M M Monwar and M L Gavrilova ldquoMultimodal biometricsystem using rank-level fusion approachrdquo IEEE Transactions onSystems Man and Cybernetics Part B Cybernetics vol 39 no4 pp 867ndash878 2009

[14] A Kumar and S Shekhar ldquoPersonal identification using multi-biometrics rank-level fusionrdquo IEEE Transactions on SystemsMan and Cybernetics Part C Applications and Reviews vol 41no 5 pp 743ndash752 2011

[15] S D Thepade and R K Bhondave ldquoNovel multimodal iden-tification technique using iris amp palmprint traits with variousmatching score level proportions using BTC of bit plane slicesrdquoin Proceedings of the International Conference on PervasiveComputing (ICPC rsquo15) pp 1ndash4 Pune India January 2015

[16] B Yang C Busch K de Groot H Xu and R N J VeldhuisldquoPerformance evaluation of fusing protected fingerprint minu-tiae templates on the decision levelrdquo Sensors vol 12 no 5 pp5246ndash5272 2012

[17] A K Jain F Patrick andA Ross ArunHandbook of BiometricsSpringer Berlin Germany 2008

[18] V Conti G Milici P Ribino F Sorbello and S VitabileldquoFuzzy fusion in multimodal biometric systemsrdquo inKnowledge-Based Intelligent Information and Engineering Systems 11thInternational Conference KES 2007 XVII Italian Workshop onNeural Networks Vietri sul Mare Italy September 12-14 2007Proceedings Part I vol 4692 of Lecture Notes in ComputerScience pp 108ndash115 Springer Berlin Germany 2007

[19] W Yang J Hu S Wang and C Chen ldquoMutual dependency offeatures in multimodal biometric systemsrdquo Electronics Lettersvol 51 no 3 pp 234ndash235 2015

[20] F Besbes H Trichili and B Solaiman ldquoMultimodal biometricsystem based on fingerprint identification and iris recogni-tionrdquo in Proceedings of the 3rd International Conference onInformation and Communication Technologies From Theory toApplications (ICTTA rsquo08) pp 1ndash5 IEEE Damascus Syria April2008

[21] M Lades J C Vorbrueggen J Buhmann et al ldquoDistortioninvariant object recognition in the dynamic link architecturerdquoIEEE Transactions on Computers vol 42 no 3 pp 300ndash3111993

[22] T Ojala M Pietikainen and D Harwood ldquoA comparativestudy of texture measures with classification based on featuredistributionsrdquo Pattern Recognition vol 29 no 1 pp 51ndash59 1996

[23] Y Raja and S Gong ldquoSparse multiresolution local binarypatternsrdquo in Proceedings of the 17th British Machine VisionConference Edinburgh UK September 2006

[24] S Liao X Zhu Z Lei L Zhang and S Z Li ldquoLearning multi-scale block local binary patterns for face recognitionrdquo inProceedings of the International Conference on Biometrics (ICBrsquo07) pp 828ndash837 Seoul Republic of Korea August 2007

[25] Z Guo L Zhang D Zhang and X Mou ldquoHierarchical mul-tiscale LBP for face and palmprint recognitionrdquo in Proceedings

of the 17th IEEE International Conference on Image Processing(ICIP rsquo10) vol 17 pp 4521ndash4524 IEEE Hong Kong September2010

[26] X Zhou and B Bhanu ldquoFeature fusion of side face and gait forvideo-based human identificationrdquo Pattern Recognition vol 41no 3 pp 778ndash795 2008

[27] B V Kumar A Mahalanobis and R D Juday Correlation Pat-tern Recognition Cambridge University Press Cambridge UK2005

[28] K Nandakumar Y Chen A K Jain and S C Dass ldquoQuality-based score level fusion in multibiometric systemsrdquo in Proceed-ings of the 18th International Conference on Pattern Recognition(ICPR rsquo06) vol 4 pp 473ndash476 IEEE Hong Kong August 2006

[29] Y Zhang D Sun and Z Qiu ldquoHand-based feature level fusionfor single sample biometrics recognitionrdquo in Proceedings ofthe 1st International Workshop on Emerging Techniques andChallenges for Hand-Based Biometrics (ETCHB rsquo10) pp 1ndash4Istanbul Turkey August 2010

[30] H Korves L Nadel H Korves H Nadel B Ulery and DMasildquoMulti-biometric fusion from research to operationsrdquo Sigmamitretek systems pp 39ndash48 Summer 2005

[31] G Aguilar G Sanchez K Toscano M Nakano and H PerezldquoMultimodal biometric system using fingerprintrdquo in Proceed-ings of the International Conference Intelligent Advanced Sys-tem (ICIAS rsquo07) pp 145ndash150 IEEE Kuala Lumpur MalaysiaNovember 2007

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



Shekhar [14] investigated multiple palmprint recognitionusing rank level fusion which uses borda count bucklinhighest rank and logistic regressionMatch score level fusionusing feed forward neural network for the fusion of face andpalmprint has been investigated by Thepade and Bhondave[15] Yang et al [16] investigated multiple dependency ofpalmprint using feature level fusion and score level fusionWang and Han [9] investigated face iris fusion using scorelevel fusion Jain et al [17] investigated the multimodalbiometric system based on the face and hand geometrybiometrics at the score level

Conti et al [18] proposed a multimodal biometric basedon two-fingerprint acquisition which uses score level fusionand obtained an improvement of 6 when compared tomonomodal biometric based system Yang et al [19] usedmatched score level fusion to fuse hand geometry fingerprintand palmprint multimodal biometric system He used aself-constructed database of 97 subjects Besbes et al [20]proposed a hybrid multitrait biometric method using irisand fingerprint Yang et al [16] proposed decision levelfusion fingerprint templates Here assessment was taken byindividual unimodal assessment through an ldquoANDrdquo operator

Most significant contribution published in recent yearspertaining to multimodal biometric fusion focused exten-sively on fusing data at the matching score level and decisionlevel It has been observed thatmost of the important featuresare lost on performing data fusion at the latter stages (matchscore level and decision level) In spite of the abundanceof investigations related to multimodal biometrics relativelylittle work was done at feature level fusion since featurefusion has rich information content compared to fusionat the later stages Therefore the current exploration on amultimodal biometric fusion at the feature level is anticipatedto attain improved recognition accuracy compared to thefusion at the later stages

2 Feature Fusion Using HierarchicalMultiresolution LBP and Gabor

This research mainly discusses the multifeature fusion ofpalmprint and iris biometrics using feature level fusion HereFigure 1 illustrates feature fusion using hierarchical multires-olution LBP and Gabor It consists of three major blockspreprocessing feature extraction and fusion Multimodalmultifeature-based biometric system involves the followingsteps

(i) The two modalities presumed are palmprint and irisimage which are given as input

(ii) The Gabor feature and hierarchical multiresolutionLBP features of palmprint and iris image respectivelyare taken

(iii) Images are fused by using feature level fusion(iv) 119870 nearest neighbor is used for classification(v) Recognition accuracy is calculated

21 Gabor Wavelets Gabor wavelets are a filter bank con-sisting of Gabor filters with diverse scales and rotation It is

Registration

Identificationverification





Feature level fusion


Matching

Database


Figure 1 Feature fusion using hierarchical multiresolution LBP andGabor

efficient for analyzing dissimilar phased features like abruptridges or edges Gabor space is extremely useful in variousmedical image-processing purposes (Lades et al [21])Mainlythe Gabor wavelets were developed to represent the receptivefields of simple cells in the visual cortex However in practicethey confine to most of the salient properties togetherwith frequency selectivity orientation selectivity and spatiallocalization Actually here the image is conlvolved with abank of Gabor filters of different orientations and scalesGabor wavelet has the following general form as in

1205951205831]1(119911)

=

100381710038171003817100381710038171198701205831]1

10038171003817100381710038171003817

2

1205902119890minus1198701205831]1 2119885221205902

[1198901198951198701205831]1119885 minus 119890

minus12059022]

(1)

where represents the norm operator ]1 and 1205831 are scaleand the orientation respectively of the Gabor kernel 119911 =

(1199091 1199101) represents a variable in spatial domain and 1198701205831]1

represents wave vector and 120590 is the standard deviationThe wave vector is represented in

1198701205831]1

= 119870]1

(cos1206011205831

+ 119895 sin1206011205831

) (2)

where 119870]1

= 119870max119891]1

and 1206011205831

= 1205831205878 with 119870max being themaximum frequency and 119891 is the spacing factor In thisresearch the Gabor kernel filter used is of three differentscales and four orientations Figure 2 shows the Gabor kernelfilter output

22 Hierarchical Multiresolution Local Binary Pattern Ojalaet al [22] introduced local binary pattern in 1996 The localbinary pattern is a gray scale invariant texture measureand is a helpful tool to model texture images It tags thepixels representation by using threshold of the pixels ofthe local neighbor around each pixel and considers theresult as binary numbers It is a combining approach todivergent statistical and structural forms of texture anal-ysis The major property of hierarchical multiresolutionLBP is its robustness to monotonic gray scale alterationscaused An added advantage of hierarchical multiresolutionLBP is its computational simplicity to analyze images in






LBP119872119903119870119888 =119872minus1

sum

119898

119891 (119870119899 119870119888) 2119899

119891 (119870119899 119870119888) =

1

0

119868 (119870119899) ge 119868 (119870119888)

119868 (119870119899) gt 119868 (119870119888)

(3)



(119909119899 119910119899) = [119909119888 + 119903 cos(2120587119899119872

) 119910119888 minus 119903 sin(2120587119899119872

)] (4)










Furtherprocessing

Furtherprocessing



R = 3

R = 2




R = 1

R = 2





Iris input













single vector


single vector

Matching Decision



Database

Feature fusion






(a) (b)

(c) (d)

(e) (f)






(a) (b)

(c) (d)





3 2 4 4 3 1 4 4 3 3 Class id 11 1 2 2 3 1 5 4 5 5 Class id 2







123456

789101112

131415161718

1920212223

0095

0096

0097

0098

0099

01

0101

0102

0103

0104

0105

Scor

e

0104 0106 0108 011 0112 0114 0116 01180102Data






1 1 1 4 Class id 11 2 3 4 Class id 2




0002004006008

01Fa

lse re

ject

ion

rate


EER = 00528

EER = 0042

EER = 00378



5 Conclusion




















References










































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in








LBP119872119903119870119888 =119872minus1

sum

119898

119891 (119870119899 119870119888) 2119899

119891 (119870119899 119870119888) =

1

0

119868 (119870119899) ge 119868 (119870119888)

119868 (119870119899) gt 119868 (119870119888)

(3)



(119909119899 119910119899) = [119909119888 + 119903 cos(2120587119899119872

) 119910119888 minus 119903 sin(2120587119899119872

)] (4)










Furtherprocessing

Furtherprocessing



R = 3

R = 2




R = 1

R = 2





Iris input













single vector


single vector

Matching Decision



Database

Feature fusion






(a) (b)

(c) (d)

(e) (f)






(a) (b)

(c) (d)





3 2 4 4 3 1 4 4 3 3 Class id 11 1 2 2 3 1 5 4 5 5 Class id 2







123456

789101112

131415161718

1920212223

0095

0096

0097

0098

0099

01

0101

0102

0103

0104

0105

Scor

e

0104 0106 0108 011 0112 0114 0116 01180102Data






1 1 1 4 Class id 11 2 3 4 Class id 2




0002004006008

01Fa

lse re

ject

ion

rate


EER = 00528

EER = 0042

EER = 00378



5 Conclusion




















References










































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in





Furtherprocessing

Furtherprocessing



R = 3

R = 2




R = 1

R = 2





Iris input













single vector


single vector

Matching Decision



Database

Feature fusion






(a) (b)

(c) (d)

(e) (f)






(a) (b)

(c) (d)





3 2 4 4 3 1 4 4 3 3 Class id 11 1 2 2 3 1 5 4 5 5 Class id 2







123456

789101112

131415161718

1920212223

0095

0096

0097

0098

0099

01

0101

0102

0103

0104

0105

Scor

e

0104 0106 0108 011 0112 0114 0116 01180102Data






1 1 1 4 Class id 11 2 3 4 Class id 2




0002004006008

01Fa

lse re

ject

ion

rate


EER = 00528

EER = 0042

EER = 00378



5 Conclusion




















References










































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in




(a) (b)

(c) (d)

(e) (f)






(a) (b)

(c) (d)





3 2 4 4 3 1 4 4 3 3 Class id 11 1 2 2 3 1 5 4 5 5 Class id 2







123456

789101112

131415161718

1920212223

0095

0096

0097

0098

0099

01

0101

0102

0103

0104

0105

Scor

e

0104 0106 0108 011 0112 0114 0116 01180102Data






1 1 1 4 Class id 11 2 3 4 Class id 2




0002004006008

01Fa

lse re

ject

ion

rate


EER = 00528

EER = 0042

EER = 00378



5 Conclusion




















References










































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in




(a) (b)

(c) (d)





3 2 4 4 3 1 4 4 3 3 Class id 11 1 2 2 3 1 5 4 5 5 Class id 2







123456

789101112

131415161718

1920212223

0095

0096

0097

0098

0099

01

0101

0102

0103

0104

0105

Scor

e

0104 0106 0108 011 0112 0114 0116 01180102Data






1 1 1 4 Class id 11 2 3 4 Class id 2




0002004006008

01Fa

lse re

ject

ion

rate


EER = 00528

EER = 0042

EER = 00378



5 Conclusion




















References










































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in





123456

789101112

131415161718

1920212223

0095

0096

0097

0098

0099

01

0101

0102

0103

0104

0105

Scor

e

0104 0106 0108 011 0112 0114 0116 01180102Data






1 1 1 4 Class id 11 2 3 4 Class id 2




0002004006008

01Fa

lse re

ject

ion

rate


EER = 00528

EER = 0042

EER = 00378



5 Conclusion




















References










































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in





















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 performance evaluation of multimodal...

Documents