Research Article Comparative Study of Multimodal Biometric...

Research ArticleComparative Study of Multimodal Biometric Recognition byFusion of Iris and Fingerprint

Houda Benaliouche and Mohamed Touahria

Computer Science Department University of Ferhat Abbas Setif 1 Pole 2 - El Bez 19000 Setif Algeria

Correspondence should be addressed to Houda Benaliouche houdaaimargmailcom

Received 28 August 2013 Accepted 17 November 2013 Published 29 January 2014

Academic Editors J Shu and F Yu

Copyright copy 2014 H Benaliouche and M Touahria 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 investigates the comparative performance from three different approaches for multimodal recognition of combinediris and fingerprints classical sum rule weighted sum rule and fuzzy logic method The scores from the different biometric traitsof iris and fingerprint are fused at the matching score and the decision levels The scores combination approach is used afternormalization of both scores using themin-max rule Our experimental results suggest that the fuzzy logicmethod for thematchingscores combinations at the decision level is the best followed by the classical weighted sum rule and the classical sum rule in orderThe performance evaluation of each method is reported in terms of matching time error rates and accuracy after doing exhaustivetests on the public CASIA-Iris databases V1 and V2 and the FVC 2004 fingerprint database Experimental results prior to fusionand after fusion are presented followed by their comparison with related works in the current literature The fusion by fuzzy logicdecision mimics the human reasoning in a soft and simple way and gives enhanced results

1 Introduction

Biometrics refers to identity verification of persons accordingto their physical or behavioral characteristics Many physicalbody parts andpersonal features have beenused for biometricsystems fingers hands feet faces irises retinas ears teethveins voices signatures typing styles gaits odors and DNAPerson verification based on biometric features has attractedmore attention in designing security systems [1] Howeverno single biometrical feature can meet all the performancerequirements in practical systems [2] Most of biometricsystems are far from satisfactory in terms of user confidenceand user friendliness and have a high false rejection rate FRRThere is a need for development of novel paradigms andprotocols and improved algorithms for human recognitionUnimodal biometric systems use one biometric trait torecognize individuals These systems are far from perfectand suffer from several problems like noise nonuniversalitylack of individuality and sensitivity to attack Multimodalbiometric systems use multiple modalities to overcome thelimitations that arise when using single biometric trait torecognize individuals Multiple biometric systems perform

better than unimodal biometric systems The use of only onebiometric trait susceptible to noise bad capture and otherinherent problems makes the unimodal biometric systemunsuited for all applications

Many works in the literature have demonstrated that thedrawbacks of the unimodal biometric systems are mainlygenuine and imposters identification failure due to theintraclass variations and the interclass similarities whilethe drawbacks associated with multimodal biometrics areincreased complicity of the system with two or more sensors[2ndash6] and thus higher cost as well as inconvenience of usingseveral biometrics So identification of person with highaccuracy and less complexity of the system is becomingcritical in a number of security issues in our society Irisand fingerprint biometrics are more simple accurate andreliable as compared to other available traitsThese propertiesmake their fusion particularly promising solution to theauthentication problems today Moreover fusion of iris andfingerprint is more reliable than fusion of each one withanother biometric like face [7] However iris biometric hasmore features and stability and resistance to attacks thanfingerprint biometric despite this the conventional fusion

Hindawi Publishing Corporatione Scientific World JournalVolume 2014 Article ID 829369 13 pageshttpdxdoiorg1011552014829369

2 The Scientific World Journal

methods still use the same weight in fusion for each singlebiometric and this is the reason for why their best error ratesare far from perfect False accept rate identifies the number oftimes an imposter is classified as a genuine user by the systemand false reject rate pertains to misidentification of a genuineuser as an imposter Although ideally both FAR and FRRshould be as close to zero as possible in real systems howeverthis is not the case [8] For an ideal authentication systemFAR and FRR indexes are equal to 0 To increase the relatedsecurity level system parameters are then fixed in order toachieve the FAR = 0 point and a corresponding FRR point[9]

In this paper a novel combination of iris and fingerprintbiometrics is presented in order to achieve best compromisebetween a zero FAR and its corresponding FRR in ourapproach iris trait has more weight in fusion with fingerprintand the system decision is made to have more intermediatevalues between bad and good recognition the weight issimply an appreciation we assign to thematching distance foreach single biometric set by fuzzy membership function andwe use major concepts of fuzzy logic introduced by Zadeh[10] which are fuzzy sets fuzzy membership function andfuzzy inference system The fuzzy inference system mimicsour human thinking and this is mainly the reason we getenhanced results

The objective of this research is threefold first designingand implementing monomodal systems for the biometricrecognition of iris and fingerprint these systems will servelatter for comparison second designing and implement-ing a multimodal biometric system of combined iris andfingerprint using the previous monomodal systems withthree different matching algorithms two classical matchingalgorithms and our proposed one based on fuzzy logicthird carrying out exhaustive and intensive tests on the irisand fingerprint databases using the proposed recognitionschemes to conclude at the end the best one At last acomparison of the achieved results with similar works in thecurrent literature is given

The paper is organized as follows in the next sectionrelated works are presented followed by a presentation ofstate of the art of multimodal biometric in Section 3 thework methodology is presented and the two modalities arecombined through three different experiments using twolevels of fusion one at the score level and the other atthe decision level in Section 4 the details of the systemimplementation is given and the databases involved in thework are presented the experimental results prior to fusionand after fusion are presented in Section 5 followed by theirevaluation and comparison in Section 6 Conclusion is givenin the last section

2 Related Works

Multimodal biometrics has been proposed by Ross and Jainin 2003 [11]The concept of biometric multimodalities fusionis introduced with different fusion strategies and variouslevels of fusion are also presented [2 4 6 12ndash17] Fusionof iris and fingerprint has attracted a lot of attention andresearchers have presented variety of approaches in the

literature [7 8 16 18 19] Baig et al [8] in 2009 proposed aframework formultimodal biometric fusion based on utiliza-tion of a single matcher implementation for both modalities(iris andfingerprint) For their experiment they used theWestVirginia Universityrsquos multimodal database containing 400images (4 enrolment images times 100 users) and the threshold isset to the equal error rate EERThe comparison is beingmadein terms of percentage improvement in EER rather than theEER values themselves Jagadeesan et al [16] in 2010 intro-duced a technique for cryptographic key generation by fusingfingerprint and iris biometrics The fingerprint extractor isminutia based while the iris extractor is based on canny edgedetector and Hough transform (Daugmanrsquos approach) Theminutiae points and texture properties were first extractedfrom fingerprint and iris images respectively and then theywere fused at the feature level to obtain the multibiometrictemplate and subsequently a 256-bit secure cryptographic keyfrom the multibiometric template is generated

In 2011 Jameer Basha et al [18] introduced a newframework for iris and fingerprint fusion at rank levelthey conducted experimental tests using three implementedfusion methods highest rank method Borda count methodand logistic regression method Their work achieved the bestexecution time required to match which is equal to 045seconds for the highest rank method with optimal FAR andFRR equal respectively to 0 and 025

In 2012 Radha and Kavitha [19] presented a novel fusionscheme of fingerprint and iris modalities at feature extractionlevel The scheme uses a concatenated feature vector fromboth iris and fingerprintThe logGabor filter is used to extractthe feature vectors of both modalities then the phase datafrom 1D log Gabor filters is extracted and quantized to fourlevels to encode the unique pattern of iris andFingerprint intobitwise biometric template Hamming distance (HD) is usedto generate a final match score Experimental results wereverified on database of 50 users accounting to FAR = 0 andFRR = 43The execution time required tomatch is reducedto 014 seconds

In 2013 Abdolahi et al [7] presented a multimodalbiometric system (fingerprint and iris) using fuzzy logic andweighted code After converting fingerprint and iris imagesto a binary code the decision level fusion is used to combinethe results Fingerprint code is weighed as 20 and iris codeas 80 The work achieved 2 FAR and FRR and 983accuracy

3 State of the Art of Multimodal Biometric

In this section we summarize the main ideas and principlesinvolved in the area of multimodal biometric recognition

31 Multimodal Biometrics versus Multibiometrics As ex-plained by most research papers in the field of biometricrecognition [5 12 16 20] the term ldquomultimodal biometricrdquorefers to multiple biometric traits used together at a specificlevel of fusion to recognize persons The ldquomultibiometricsrdquoincludes either the use of multiple algorithms also calledclassifiers at enrolment or matching stages for the samebiometric trait or the use of multiple sensors of the same

The Scientific World Journal 3

Multiplealgorithms

sensors

Repeatedinstances

Multiplemodalities

Scenariosof fusion

Multiple Multipleinstances

Figure 1 Fusion scenarios of multimodal systems

biometric trait like using different instruments to capture thebiometric details or using multiple instances of the samebiometric trait like the use of fingerprints of three fingers orfinally using repeated instances like repeated impressions ofone finger

32 Fusion in Biometry In order to join two or morebiometric traits a method called ldquofusionrdquo is used [12] Fusionin biometry refers to the process of combining two or morebiometric modalities In this section we present the differentscenarios of fusion used by multimodal biometric systems Itis worth noting that the multimodality does not involve theuse of multiple biometric modalities in the strict sense of theterm (such as combining iris and fingerprint) but itsmeaningis broader as defined in the following by the various scenariosof fusion (see Figure 1)

321 Level of Fusion Five levels of fusion in multimodalsystems were introduced in the literature [4 12] which are thefollowing

(1) Sensor Level Multisensorial biometric systems sample thesame instance of a biometric trait with two or more distinctlydifferent sensors [14] Processing of the multiple samples canbe done with one algorithm or combination of algorithmsExample face recognition application could use both a visiblelight camera and an infrared camera coupled with specificfrequency

(2) Feature Level The feature level fusion is useful in classifi-cation [14] Different feature vectors are combined obtainedeither with different sensors or by applying different featureextraction algorithms to the same raw data [21]

(3) Decision Level With this approach each biometric sub-system completes autonomously the processes of featureextractionmatching and recognitionDecision strategies areusually of Boolean functions where the recognition yields themajority decision among all present subsystems [9]

(4) Rank Level Instead of using the entire template partitionsof the template are used Ranks from template partitions are

Sensorlevel

Featurelevel

Rank

Decision Score

level

levellevel

Fusion levels

Figure 2 Levels of fusion in multimodal biometric systems

consolidated to estimate the fusion rank for the classification[18] Rank level fusion involves combining identificationranks obtained from multiple unimodal biometrics It con-solidates a rank that is used for making final decision [19]

(5) Score Level It refers to the combination ofmatching scoresprovided by the different systemsThe score level fusion tech-niques are divided into two main sets fixed rules (AND ORmajority maximumminimum sum product and arithmeticrules) and trained rules (weighted sum weighted productfisher linear discriminate quadratic discriminate logisticregression support vector machine multilayer perceptronsand Bayesian classifier ) [22] Figure 2 shows the five levels ofbiometric fusion

322 Normalization Score normalization brings bothmatching scores between 0 and 1 [23] The normalization ofboth scores by the min-max rule are given by

119873Iris =MSIris minusminIrismaxIris minusminIris

(1)

119873Finger =MSFinger minusminFingermaxFinger minusminFinger

(2)

where MSIris and MSFinger are the matching scores obtainedfrom iris and fingerprint modalities respectively minIrisand maxIris are the minimum and maximum scores for irisrecognition andminFinger andmaxFinger are the correspondingvalues obtained from fingerprint trait Other normalizationalgorithms also exist like119885-score TanH and Sigmoid whichgave very good results TanH method gave the best resultbut it involved a lot of parameters Z-score and min-max aresimple but they are insensitive to the presence of outliers [17]

4 The Research Methodology

Figure 3 shows the different stages included in our mul-timodal recognition system and the overall system designshows the following

(i) The level at which the biometric information of theiris and fingerprint are fused is indicated (here twolevels are used the score level fusion is used for the


Iris DB

Matching algorithm 1

Irisinput

EnhancementFeature

extractionmodule

Matchingmodule

Iris score

Decision Fusion module Normalisation

Fusionchoice

FuzzificationFuzzy logic

fusionmodule

DefuzzificationDecision

Fingerprintinput

Enhancement Fingerprintscore


Fingerprint

DB

Featureextraction

module

Matchingmodule

Figure 3 Flow chart of the application showing the main modules of the multimodal biometric recognition system

classical fusion and the decision level fusion is usedfor the fusion with fuzzy logic)

(ii) The fusion approach used is the approach by combin-ing scores when the method of fusion is classic

(iii) The other fusion approach used is fusion of decisionswhen the method of fusion is fuzzy

(iv) The normalization of scores is required prior to thefusion only for the classical fusion (which is explainedby the use of the approach by combining scores forboth classical sum rulematching andmatching by thelinear weighted sum rule)

(v) Fusion by fuzzy logic does not require normalizationof scores and only decisions are used by the fuzzyinference system

(vi) Threematching algorithms are used the classical sumrule matching the weighted sum rule matching andour proposed matching with fuzzy logic

The conventional fusion methods [2ndash4 9 11ndash17] usethe same weight for each single biometric trait but somebiometric traits are more precise than the other ones theyhave more stability and resistance to attacks So in ourapproach iris trait hasmoreweight in fusionwith fingerprintWeight here is not a number assigned to the matchingscore but a decision with intermediate values related to thematching distance

In this work we have implemented two different architec-tures of the combined iris and fingerprint biometric recog-nition system in order to compare the recognition results

Iris sensor Fingerprintsensor

Iris classifier Fingerprintclassifier

Iris score Fingerprintscore

Fusion ofscores

Threshold

Comparison

Result

Figure 4 Score level fusion using iris and fingerprint biometricmo-dalities

(in terms of time accuracy and error rates) of both systemarchitectures and conclude the best one The first systemarchitecture (see Figure 4) is based on the classical fusionof scores The second system architecture (see Figure 5) isbased on our proposed fuzzy logic matching using iris andfingerprint decisions


41 Classical Matching Strategies

411 Hamming Distance BasedMatching For the iris modal-ity we use the hamming distance based matching

HD =sum119873

119895=1119883119895 (XOR) 119884119895 (AND) 119883119899

1015840

119895(AND) 1198841198991015840

119895

119873 minus sum119873

119896=1119883119899119896 (OR) 119884119899119896

(3)

The hamming distance HD is calculated using formula(3) where 119883

119895and 119884

119895are the models to compare bit by bit

119883119899119895and119884119899

119895are the noise masks for 119883

119895and 119884

119895 and119873 is the

number of bits represented by eachmodel For the fingerprintmodality we use the Euclidian distance based matching (seeformula (4))

ED = radic119899

sum

119894=1

(119883119894minus 119884119894)

2

(4)

where119883119894and 119884

119894are the models to compare

412 The Sum Rule Based Matching After the normalizationof both iris and fingerprint scores the score of fusion iscalculated as presented by formula (5)

1198781015840=

119899

sum

119896=1

119878 (5)

where 1198781015840 is the score of fusion and 119899 is the number of thescores here 119899 = 2

413 The Weighted Sum Rule Based Matching After thenormalization of fingerprint and iris scores the score offusion is calculated as presented by formula (6)

1198781015840= 1205721198781+ (1 minus 120572) 1198782

(6)

where 1198781015840 is the score of fusion 1198781and 119878

2are respectively the

scores of the biometric modalities to be combined and120572 isthe weight assigned to each modality

In our experimentation we set 120572 to 08 for the irismodality and 1 minus 120572 = 02 for the fingerprint modality

42 Fuzzy Logic BasedMatching Our proposed fuzzymatch-ing algorithm assigns a specific appreciation to each decisionaccording to the best threshold minimizing both FRR andFAR The fuzzy if-then rules produce decisions according tothe matching distance calculated for each modality

For that

(i) we define two fuzzy variables for the input ldquofingerrdquofor the fingerprint trait and ldquoirisrdquo for the iris trait

(ii) we define the output fuzzy variable ldquofusionrdquo(iii) each variable is represented by a trapezoidal fuzzy set(iv) for the inputs we define three fuzzy sets according to

the matching distance bad medium and good(v) the output is fuzzy either very bad or bad or medium

or good or very good or excellent




Result

Iris decisionFingerprint

decision

Decisionsfusion

Fuzzyinference

Fuzzy if then rules

Figure 5 Decision level fusion using iris and fingerprint biometricmodalities

Deg

ree o

f mem

bers

hip

Good Medium Bad1

0

Threshold

S1 S2 S3 S4 S5 S6

Figure 6 Fuzzy sets of the proposed entries and their trapezoidalmembership functions

As shown by Figure 6 [1198781 1198782] is the interval of thresholds

belonging to the fuzzy set ldquogoodrdquo [1198783 1198784] is the interval of

thresholds belonging to the fuzzy set ldquomediumrdquo [1198785 1198786] is the

interval of thresholds belonging to the fuzzy set ldquobadrdquoThe fuzzy if-then rules combining decisions from iris and

fingerprint modalities respect the following fuzzy rules

If (finger is bad) and (iris is bad) then (fusion is verybad)

If (finger is bad) and (iris is medium) then (fusion ismedium)

If (finger is bad) and (iris is good) then (fusion is good)If (finger is medium) and (iris is bad) then (fusion isbad)

If (finger is medium) and (iris is medium) then (fusionis good)

If (finger is medium) and (iris is good) then (fusion isvery good)

If (finger is good) and (iris is bad) then (fusion is me-dium)


If (finger is good) and (iris is medium) then (fusion isvery good)

If (finger is good) and (iris is good) then (fusion is ex-cellent)

We have set the if-then rules according to the followingcriteria

(i) the iris decision is more reliable than the fingerprintdecision so we give more weight to the iris decisionin fusion with the fingerprint decision

(ii) the fusion decision is one of the following sets verybad bad medium good very good excellent

(iii) in the cases where the iris decision is ldquobadrdquo the fusiondecision should be either ldquobadrdquo or ldquovery badrdquo orldquomediumrdquo even if the fingerprint decision is good

(iv) in the cases where the iris decision is ldquogoodrdquo thefusion decision should be either ldquogoodrdquo or ldquoexcellentrdquoeven if the fingerprint decision is ldquobadrdquo

5 System Implementation

The programming language used to implement our system isMATLAB 7100(R2010a) MATLAB as well as its interactiveenvironment is a high-level language that allows the execu-tion of tasks requiring high computing power and whoseimplementation will be much easier and faster than with tra-ditional programming languages such as C C++ It has sev-eral toolkits in particular image processing ldquoImageProcessingToolboxrdquo which proposes a set of algorithms and graphicalreference tools for the processing analysis visualization andimage processing algorithm development Our applicationis implemented on a laptop (HP630) Intel CORE I3 CUPM370 with 2Giga byte of RAM and 320Giga byte hard drivedisk HDD and has a 240GHz speedTheminimum requiredmaterial characteristics for the application are 512Mega byteof RAM and 80Giga byte hard drive To perform tests withour application we use four databases which are as follows

(i) CASIA-Iris V1 [24] CASIA V1 contains 756 imagesfrom 108 eyes For each eye 7 images are capturedin two sessions with a homemade iris camera wherethree samples are collected in the first session and fourin the second session All images are stored as BMPformat with resolution 320 lowast 280

(ii) CASIA-Iris V2 [25] contains 2400 images from 120eyes For each eye 10 images are captured using twodifferent instruments (OKI and Pattek) All imagesare stored in BMP format with resolution 640 lowast 480CASIA-Iris V2 contains blurry images with differentilluminations and wearing glasses is authorized Thedatabase is available for free on demand

(iii) FVC 2004 [26] contains four sets DB1 A DB2 ADB3 A and DB4 A Each of these databases contains800 fingerprints equivalent of one hundred (100)individuals each having eight (08) impressions FVC2004 database is characterized by different fingerprintimage qualities The database is purchased uponrequest

Figure 7 GUI of the verification process in the iris monomodalrecognition system

(iv) Our proposed database of combined irises and fin-gerprints made from an equivalent number of irisesfrom CASIA-Iris V2 database and fingerprints fromFVC2004 database (50 subjects lowast 10 images)

For a given database if 119888 represents number of classes and119899 represents total number of images per class then intraclasscombinations are calculated as (119899minus1times(1198992)times119888) and interclasscombinations are calculated as (119888 times (119888 minus 1) times 119899 times 119899) [27] Forexample for CASIA V1 database intra-class combinationsare worked out as ((7 minus 1) times (72) times 108) and inter-classcombinations are worked out as (108 times 107 times 7 times 7)

For the database CASIA-V2 the intra-class combinationsare worked out as ((20 minus 1) times (202) times 120) = 22800 andinterclass are worked out as class combinations (120 times 119 times20 times 20) = 5712000

For database FVC 2004 intra-class combinations areworked out as ((800minus1) times (8002) times 4) = 1278400 and inter-class combinations are worked out as (4 times 3 times 800 times 800) =7680000

6 Experimental Results

The application is divided mainly into three modules

61 Iris Recognition Module Both verification and identifi-cation processes are implemented Figures below present thegraphical user interfaces GUIs allowing the user to load aniris image from a database and to do segmentation featureextraction and either verification (see Figure 7) (the user hasto upload another iris image) or identification (see Figure 8)(the system searches similar code in database)

62 Fingerprint RecognitionModule Like the iris recognitionmodule both verification and identification processes areimplemented Figure 9 shows the GUI allowing the user toload two fingerprint images and then visualize the results ofeach step of the fingerprint recognition algorithm (binarisa-tion region of interest and the orientation field localisationthe process of image thinning also called skeletonization theextraction of minutia the elimination of false minutia andfinally the matching by the Euclidian distance)


Figure 8 GUI of the identification process in the iris monomodalrecognition system

Figure 9 GUI of the verification process in the fingerprint mono-modal recognition system

The identification process in the fingerprint monomodalrecognition system consists of matching the generated codefrom the input image with all codes stored in databases if theidentification failed the user is asked either to add or not thenonidentified image to a chosen database (see Figure 10)

63 Combined Iris and Fingerprint Recognition ModuleThree matching algorithms are implemented first is thematching using the fusion of iris and fingerprint by thesum rule second is the matching of both modalities by theweighted sum rule and the final is the matching using thefusion by the fuzzy logic if-then rules and the fuzzy inferencesystem

Figure 11 shows the GUI allowing the user to see theverification result of iris and fingerprint combined biometrictraits

Figure 12 presents the graphical user interface of therecognition module based on the fusion by the weightedsum rule and the score normalization is done prior to fusionusing the min-max rule and then the fusion is done in ourexperimentation we set 120572 to 08 for the iris modality and1 minus 120572 = 02 for the fingerprint modality

Figure 13 presents the graphical user interface allowingthe user to verify the similarity between two individuals byopening the fingerprint and iris images belonging to each

Figure 10 GUI of the identification process in the fingerprintmonomodal recognition system

Figure 11 GUI showing the matching using the fusion by the sumrule

Figure 12 GUI showing the matching using the fusion by theweighted sum rule

Figure 13 GUI showing the matching using the fusion by the fuzzyinference system


individual doing feature extraction andmatching operationsbetween the two irises and the two fingerprints output thematching distances and the decisions of both modalities andthen plot the fuzzy membership function for each decisionand finally calculate the decision of the combined modalitiesand plot its fuzzy membership function

7 Performance Evaluation and Comparison

In order to test our proposed schemes for monomodaland multimodal biometric recognition systems and proceedwith their evaluation and comparison we do the followingexperiments

Experiment 1 Both verification and identification processesare implemented within a monomodal iris recognition sys-tem We use the public code of Masek and Koveski [28] forthe verification andwe extend it to perform the identificationThe feature extractor employed for Iris modality is based onDaugmanrsquos approach [29] and was implemented by Masekand Koveski [28] An Iris code comprising bit streams calledIriscode by Daugman is generated The hamming distancebased matcher provides the matching score The experimentuses CASIA-V1 iris database

Experiment 2 Both verification and identification processesare implemented within a monomodal iris recognition sys-tem Like Experiment 1 we use the public code of Masekand Koveski [28] for the verification and we extend it toperform the identification The experiment uses CASIA-V2iris database

Experiment 3 Both verification and identification processesare implemented within a monomodal fingerprint recog-nition system and we propose a minutia based fingerprintrecognition system using the algorithm of Jagadeesan et al[16] to localize the region of interest and the orientation fieldand the algorithm of Jain et al [30] for the extraction ofminutiae and posttreatment Matching is based on EuclidiandistanceThe experiment uses FVC2004 fingerprint database

Experiment 4 Only verification process is implementedwithin a multimodal biometric recognition system of com-bined iris and fingerprint using the sum rule basedmatchingThe experiment uses an equivalent number of images fromCASIA Iris-V2 and FVC2004 fingerprint databases (5 fromeach modality lowast 50 subjects)

Experiment 5 Only verification process is implementedwithin a multimodal biometric recognition system of com-bined iris and fingerprint using the weighted sum rulebased matching The experiment uses an equivalent numberof images from CASIA Iris-V2 and FVC2004 fingerprintdatabases (5 from each modality lowast 50 subjects)

Experiment 6 Only verification process is implementedwithin a multimodal biometric recognition system of com-bined iris and fingerprint using our proposed fuzzy logicbased matching The experiment uses an equivalent number

Table 1 Matching time comparison

Verification IdentificationIris recognition using CASIA Iris-V1 0138 01797Iris recognition using CASIA-Iris-V2 0155 0298Fingerprint recognition using FVC 2004 0087 015876Fusion by sum rule 0256 Fusion by weighted sum rule 02487 Proposed fusion by fuzzy logic 01754

of images from CASIA Iris-V2 and FVC2004 fingerprintdatabases (5 from each modality lowast 50 subjects) For all thetests we use the FVC2004 testing protocol [26] for thefingerprint and iris recognition modules

The fingerprint testing protocol is described as follows

(i) Genuine recognition attempts the template of eachimpression is matched against the remaining impres-sions of the same individual but avoiding symmetricmatches

(ii) Impostor recognition attempts the template of thefirst impression is matched against the first impres-sion of the remaining individuals but avoiding sym-metric matches

The iris testing protocol is described as follows

(i) First the database is divided into two parts 40 of thedatabase is reserved to enrolment in order to estimatethe classifier parameters and 60 of the database isused to test and validate the classifier

(ii) Genuine recognition attempts the template of eachiris ismatched against the remaining irises of the sameindividual but avoiding symmetric matches

(iii) Impostor recognition attempts the template of thefirst iris is matched against the first iris of the remain-ing individuals but avoiding symmetric matches

For experiments using fusionmodule tests are conductedon a set of images belonging to 50 subjects having fivefingerprint images from FVC 2004 fingerprint database andfive iris images from CASIA-Iris V2 database

71 Time Execution Comparison The values presented inTable 1 are results of execution time using CASIA V1 CASIAV2 and FVC 2004 databases and MATLAB 7100(R2010a)programming tool

We note that the fastest system in terms of matchingrecognition time is the monomodal fingerprint recognitionsystem and this is mainly due to the use of the Euclideandistance in the matching phase (see Table 1)

Our experimental results shows that the fuzzy logicmethod for the matching scores combinations at the decisionlevel is the best in terms of matching time followed by theclassical weighted sum rule and the classical sum rule inorder


Table 2 Best FAR FRR and corresponding threshold forExperiment 1

Threshold FAR () FRR ()020 0000 99047025 0000 82787030 0000 37880035 0000 5181040 0005 0238045 7599 0000050 99499 0000



Table 4 Best FAR FRR and corresponding threshold usingExperiment 3

Threshold FAR () FRR ()10 0000 9920 0005 958030 010 702940 060 567850 10 308960 1343 287870 1395 267780 1401 159890 6787 5076100 9089 1267

72 Results in Terms of Error Rates FRR FAR and EERThe false reject rate (FRR) also known as type I errormeasures the probability of an enrolled individual not beingidentified by the system The false accept rate (FAR) alsoknown as type II error measures the probability of anindividual being wrongly identified as another individual[28] According to the statistical analysis in which we havecalculated the inter-class and the intra-class thresholds usingthe above experiments whose values minimize the rates offalse acceptance and false rejection we have estimated thebest thresholds for minimal error rates for each experimentSee Tables 2 3 4 5 and 6 (values in bold are best FAR andFRR for corresponding threshold)

Unlike Experiments 1ndash5 we have no thresholds inExperiment 6 using our proposed fuzzy matching fusion


Threshold FAR () FRR ()10 0000 9930110 008 9356120 090 6089130 27 6058140 10 2689150 1283 2578160 2095 2477170 4501 1098180 7087 960185 9281 167190 9998 000



algorithm but we have decisions Table 7 shows an exampleof its intra-class and inter-class distributions

According to our proposed fusion by fuzzy matchingscheme based on if-then rules explained earlier the resultsare either excellent or very good or good or medium or bador very bad The decision ldquomediumrdquo means that

(i) either the fingerprint recognition result is ldquobadrdquo andthe iris recognition result is ldquomediumrdquo

(ii) or the fingerprint recognition result is ldquogoodrdquo andthe iris recognition result is ldquobadrdquo If we accept thedecision ldquomediumrdquo as being genuine recognition ofthe individual so we achieve FAR = 016 and FRR =00 If we reject the decisions ldquomediumrdquo as beingimposter attempts sowe achieve FAR= 00 and FRR =005

Experimental results show that the equal error ratecalculated by Experiment 6 (our proposed fuzzy matchingfusion) is EER = 0038

Figure 14 represents the plot of FAR and FRR usingExperiment 4 (iris and fingerprint fusion based sum rulematching)

As mentioned by Figure 14 the equal error rate forExperiment 4 is 155 Figure 15 represents the plot of FAR and


Table 7 Example of intraclass and interclass distributions using Experiment 6 (our proposed fuzzy matching fusion)

Individual 1 Individual 3

Individual 1 Good Good Medium Bad Bad BadVery good Good Medium Very bad Very bad Bad

Individual 2 Very bad Bad Very bad Very bad Medium BadBad Bad Bad Bad Bad Bad

Individual 3 Bad Bad Bad Good Good GoodBad Very bad Very bad Medium Good Good

Individual 4 Bad Bad Bad Bad Bad BadBad Bad Very bad Very bad Very bad Bad

Individual 5 Bad Medium Bad Very bad Bad BadBad Bad Very bad Bad Bad Bad

Table 8 Equal error rate comparison

Experiment EERExperiment 1 iris (CASIA-Iris V1) 040Experiment 2 iris (CASIA-Iris V2) 045Experiment 3 fingerprint (FVC 2004) 05Experiment 4 iris + fingerprint (sum rule fusion basedmatching) 155

Experiment 5 iris + fingerprint (weighted sum rulefusion based matching) 083

Experiment 6 iris + fingerprint (fuzzy logic fusionbased matching) 0038

100

80

60

40

20

0

1 12 14 16 18 19

FAR ()FRR ()

Figure 14 FAR and FRR using Experiment 4 (iris and fingerprintfusion based sum rule matching)

FRR using Experiment 5 (iris and fingerprint fusion basedweighted sum rule matching)

Experimental results show that Experiment 5 achieves anequal error rate of 083 Here we compare the equal error ratesof all the experiments we have carried out (see Table 8)

Table 8 presents an equal error rate comparison of thedifferent recognition methods we have implemented we see

100

90

80

70

60

50

40

30

20

10

0

05 075 08 085 09 1

FAR ()FRR ()

Figure 15 FAR and FRR using Experiment 5 (iris and fingerprintfusion based weighted sum rule matching)

that Experiment 6 performing iris and fingerprint fusionby our proposed fuzzy logic matching method is the bestfollowed by theweighted sum rule fusion basedmatching andfinally the sum rule fusion based matching

73 Results in Terms of Accuracy Table 9 presents accuracycomparison of all the experiments we have conducted

In biometry the system accuracy is calculated as follows

AC = 100 minus FRR + FAR2

(7)

According to the results presented inTable 9 we concludethat the accuracy of the method of the fusion of decisionsby fuzzy logic is better than that of the other techniquesThis comparison is done to illustrate the fact that the pro-posed system provides improved results as compared to theresults from the individual unimodal systems and the resultsfrom the implemented multimodal systems using traditionalmatching


Table 9 Accuracy comparison of all the implemented systems

Experiment Accuracy

Experiment 1 iris (CASIA-Iris V1) 9987Experiment 2 iris (CASIA-Iris V2) 979Experiment 3 fingerprint (FVC 2004) 85Experiment 4 iris + fingerprint (sum rule fusion basedmatching) 8069



Table 10 Performance comparison with related systems

Author andreference

Level offusion Database Extractor Matcher Results

Kankrale andSapkal 2012 [9]

Featureextraction

500 images of 50 subjects(from CASIA-fingerprintV5 and CASIA-iris V1)

Minutia based extractor +Daugmanrsquos iris extractor AND rule

FAR = 0FRR = 512

Match time = 356 s

Gawande et al2012 [20]

Featureextraction 500 images of 50 subjects 1D log Gabor filter for both

modalitiesHD (hamming

distance)

FAR = 0FRR = 43

Match time = 014 s

Abdolahi et al 2013[7] Decision Not given

Modified minutia basedextractor + iris extractornot given

Fuzzy rules +weighted code

FAR = FRR = 2Match time not given

Our proposedfuzzy logic basedmatching scheme

Decision500 images of 50 subject(from FVC 2004 andCASIA-Iris V2)

Minutia based extractor +Daugmanrsquos iris extractor Fuzzy if-then rules

FAR = 0FRR = 005

Match time = 01754 s

74 Comparison with RelatedWorks in the Current LiteratureTable 10 presents a comparison of the different recognitionmethods proposed and implemented in the current literaturewe see that Experiment 6 performing iris and fingerprintfusion by our proposed fuzzy logic matching method is thebetter in terms of error rates than the other presented systemsand the matching time is comparable to that of Gawandeet alrsquos [20] system

8 Conclusion

The objective of this research is the introduction of a novelmatching approach for multimodal biometric recognitionbased on fuzzy logic The biometric traits used in our workare iris and fingerprint

In this paper a novel combination of iris and fingerprintbiometrics is presented in order to achieve best compromisebetween a zero FAR and its corresponding FRR in ourapproach iris trait has more weight in fusion with fingerprintand the system decision is made to have more intermediatevalues between bad and good recognition and the weight issimply an appreciation we assign to the matching distancefor each single biometric set by fuzzy membership functionthe fuzzy inference system mimics our human thinkingand this is mainly the reason we get enhanced results Thecontribution of this research is threefold first designingand implementing monomodal systems for the biometric

recognition of iris and fingerprint second designing andimplementing a multimodal biometric system of combinediris and fingerprint using the previous monomodal sys-tems with three different matching algorithms two classicalmatching algorithms and our proposed one based on fuzzylogic and third carrying out exhaustive and intensive testson the iris and fingerprint databases using the proposedrecognition schemes to conclude at the end the best oneAt last a comparison of the achieved results with similarworks in the current literature is given and our experimentalresults are the best in terms of matching time error rates andaccuracy

The normalization of scores is required prior to the fusiononly for the classical fusion Fusion by fuzzy logic does notrequire normalization of scores only decisions are used bythe fuzzy inference system Three matching algorithms areused the classical sum rule matching the weighted sum rulematching and our proposedmatchingwith fuzzy logicThesefusion methods act on two different levels namely

(i) the score fusion level in which we implemented themethod of the classical linear sum rule of iris andfingerprint scores and method of the weighted linearsum which give weight to iris and fingerprint sores

(ii) the decision fusion level where we have designed andimplemented a fuzzy matching technique after con-verting iris and fingerprint scores to fuzzy sets (this


conversion is called fuzzification) the fuzzy inferencesystem produced fuzzy results (bad recognition orvery bad or medium or good or very good orexcellent)

Our proposed fuzzy matching algorithm assigns a spe-cific appreciation to each decision according to the bestthreshold minimizing both FRR and FAR The fuzzy if-thenrules produce decisions according to the matching distancecalculated for each modality For experiments using fusionmodule tests are conducted on a set of images belonging to50 subjects having five fingerprint images from FVC 2004fingerprint database and five iris images from CASIA-IrisV2 database Experimental results achieved best compromisebetween FRR and FAR (0 FAR and 005 FRR) withaccuracy 99975 and EER equal to 0038 andmatching timeequal to 01754s

This work allowed us to draw the following conclusions(i) The multimodal fusion gives better results than using

a single matching recognition module for iris orfingerprint

(ii) Matching with fuzzy logic provides enhanced recog-nition results followed by the classical weighted sumrule and the classical sum rule in order

This work belongs to biometric security domain It givessolution to the problem of person identification with lowererrors high accuracy and less complexity of the system Thefuzzy logic inference system used in the matching phaseis simple and robust at the same time Inperspective wesuggest that more attention should be given to the qualityenhancement of the input biometric data in order to decreasesome biometric system failures like failure to enroll andfailure to match

Conflict of Interests

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

References

[1] R Raghavendra R Ashok and G H Kumar ldquoMultimodal bio-metric score fusion using gaussian mixture model and MonteCarlomethodrdquo Journal of Computer Science andTechnology vol25 no 4 pp 771ndash782 2010

[2] J Zhou G Su C Jiang Y Deng and C Li ldquoA face andfingerprint identity authentication system based onmulti-routedetectionrdquo Neurocomputing vol 70 no 4ndash6 pp 922ndash931 2007

[3] A Ross and A K Jain ldquomultimodal biometrics an overviewrdquoin Proceedings of the 12th European Signal Processing Conference(EUSIPCO rsquo04) pp 1221ndash1224 Vienna Austria September2004

[4] H Ailisto E Vildjiounaite M Lindholm S-M Makela andJ Peltola ldquoSoft biometrics-combining body weight and fatmeasurements with fingerprint biometricsrdquoPattern RecognitionLetters vol 27 no 5 pp 325ndash334 2006

[5] S J Xie J Yang D S Park S Yoon and J Shin ldquoState of theart in biometricsrdquo in Iris Biometric Cryptosystems J Yang andL Nanni Eds InTech 2011

[6] J Yang and X Zhang ldquoFeature-level fusion of fingerprintand finger-vein for personal identificationrdquo Pattern RecognitionLetters vol 33 no 5 pp 623ndash628 2012

[7] M Abdolahi M Mohamadi and M Jafari ldquoMultimodal bio-metric system fusion using fingerprint and iris with fuzzy logicrdquoInternational Journal of Soft Computing and Engineering vol 2no 6 pp 504ndash510 2013

[8] A Baig A Bouridane F Kurugollu and G Qu ldquoFingerprintmdashiris fusion based identification system using a single hammingdistance matcherrdquo International Journal of Bio-Science and Bio-Technology vol 1 no 1 pp 47ndash58 2009

[9] R N Kankrale and S D Sapkal ldquoTemplate level concatenationof iris and fingerprint in multimodal biometric identificationsystemsrdquo International Journal of Electronics Communication ampSoft Computing Science amp Engineering pp 29ndash36 2012

[10] L A Zadeh ldquoFuzzy setsrdquo Information and Control vol 8 no 3pp 338ndash353 1965

[11] A Ross and A Jain ldquoInformation fusion in biometricsrdquo PatternRecognition Letters vol 24 no 13 pp 2115ndash2125 2003

[12] A K Jain and A Ross ldquoMultibiometric systemsrdquo Communica-tions of the ACM vol 47 no 1 pp 34ndash40 2004

[13] D R Kisku J K Sing M Tistarelli and P Gupta ldquoMultisensorbiometric evidence fusion for person authentication usingwavelet decomposition and monotonic-decreasing graphrdquo inProceedings of the 7th International Conference on Advances inPattern Recognition (ICAPR rsquo09) pp 205ndash208 February 2009

[14] K Sasidhar V L Kakulapati K Ramakrishna and K K RaoldquoMultimodal biometric systemsmdashstudy to improve accuracyand performancerdquo International Journal of Computer Scienceand Engineering Survey vol 1 no 2 pp 54ndash60 2010

[15] L Lathaa and S Thangasamy ldquoA robust person authenticationsystem based on score level fusion of left and right irises andretinal featuresrdquo Procedia Computer Science vol 2 pp 111ndash1202010

[16] A Jagadeesan T Thillaikkarasi and K Duraiswamy ldquoCryp-tographic key generation from multiple biometric modalitiesfusing minutiae with iris featurerdquo International Journal ofComputer Applications vol 2 no 6 pp 16ndash26 2010

[17] H F Liau and D Isa ldquoFeature selection for support vectormachine-based face-iris multimodal biometric systemrdquo ExpertSystems with Applications vol 38 no 9 pp 11105ndash11111 2011

[18] A Jameer Basha V Palanisamy and T Purusothaman ldquoEffi-cient multimodal biometric authentication using fast finger-print verification and enhanced iris featuresrdquo Journal of Com-puter Science vol 7 no 5 pp 698ndash706 2011

[19] N Radha and A Kavitha ldquoRank level fusion using fingerprintand iris biometricsrdquo Indian Journal of Computer Science andEngineering vol 2 no 6 pp 917ndash923 2012

[20] U Gawande S R Nair H Balani N Pawar andMKotpalliwarldquoA high speed frequency based multimodal biometric systemusing iris and fingerprintrdquo International Journal on AdvancedComputer Engineering and Communication Technology vol 1no 2 pp 66ndash73 2012

[21] R Giot and C Rosenberger ldquoGenetic programming for multi-biometricsrdquo Expert Systems with Applications vol 39 no 2 pp1837ndash1847 2012

[22] J Fierrez-Aguilar J Ortega-Garcia and J Gonzalez-RodriguezldquoFusion strategies in multimodal biometric verificationrdquo inProceedings of the IEEE Intetrnational Conference onMultimediaand Expo (ICME rsquo03) pp 5ndash8 2003


[23] A Jain K Nandakumar and A Ross ldquoScore normalization inmultimodal biometric systemsrdquo Pattern Recognition vol 38 no12 pp 2270ndash2285 2005

[24] Chinese Academy of Sciences Institute of Automation ldquoData-base of 756 greyscale eye imagesrdquo version 10 2003 httpwwwsinobiometricscom

[25] ldquoDatabaserdquo 2013 httpwwwidealtestorg[26] D Maltoni R Cappelli J L Wayman and A K Jain ldquoFVC

third fingerprint verification competitionrdquo in Proceeding of theInternational Conference on Biometric Authentication (ICBA04rsquo04) pp 1ndash7 Hong Kong July 2004

[27] A Abhyankar and S Schuckers ldquoA novel biorthogonal waveletnetwork system for off-angle iris recognitionrdquo Pattern Recogni-tion vol 43 no 3 pp 987ndash1007 2010

[28] L Masek and P KoveskiMATLAB Source Code for a BiometricIdentification System Based on Iris Patterns The University ofWestern Australia 2003

[29] J G Daugman ldquoHigh confidence visual recognition of personsby a test of statistical independencerdquo IEEE Transactions onPattern Analysis and Machine Intelligence vol 15 no 11 pp1148ndash1161 1993

[30] A K Jain L Hong S Pankanti and R Bolle ldquoAn identity-authentication system using fingerprintsrdquo Proceedings of theIEEE vol 85 no 9 pp 1365ndash1388 1997

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



methods still use the same weight in fusion for each singlebiometric and this is the reason for why their best error ratesare far from perfect False accept rate identifies the number oftimes an imposter is classified as a genuine user by the systemand false reject rate pertains to misidentification of a genuineuser as an imposter Although ideally both FAR and FRRshould be as close to zero as possible in real systems howeverthis is not the case [8] For an ideal authentication systemFAR and FRR indexes are equal to 0 To increase the relatedsecurity level system parameters are then fixed in order toachieve the FAR = 0 point and a corresponding FRR point[9]

In this paper a novel combination of iris and fingerprintbiometrics is presented in order to achieve best compromisebetween a zero FAR and its corresponding FRR in ourapproach iris trait has more weight in fusion with fingerprintand the system decision is made to have more intermediatevalues between bad and good recognition the weight issimply an appreciation we assign to thematching distance foreach single biometric set by fuzzy membership function andwe use major concepts of fuzzy logic introduced by Zadeh[10] which are fuzzy sets fuzzy membership function andfuzzy inference system The fuzzy inference system mimicsour human thinking and this is mainly the reason we getenhanced results

The objective of this research is threefold first designingand implementing monomodal systems for the biometricrecognition of iris and fingerprint these systems will servelatter for comparison second designing and implement-ing a multimodal biometric system of combined iris andfingerprint using the previous monomodal systems withthree different matching algorithms two classical matchingalgorithms and our proposed one based on fuzzy logicthird carrying out exhaustive and intensive tests on the irisand fingerprint databases using the proposed recognitionschemes to conclude at the end the best one At last acomparison of the achieved results with similar works in thecurrent literature is given

The paper is organized as follows in the next sectionrelated works are presented followed by a presentation ofstate of the art of multimodal biometric in Section 3 thework methodology is presented and the two modalities arecombined through three different experiments using twolevels of fusion one at the score level and the other atthe decision level in Section 4 the details of the systemimplementation is given and the databases involved in thework are presented the experimental results prior to fusionand after fusion are presented in Section 5 followed by theirevaluation and comparison in Section 6 Conclusion is givenin the last section

2 Related Works

Multimodal biometrics has been proposed by Ross and Jainin 2003 [11]The concept of biometric multimodalities fusionis introduced with different fusion strategies and variouslevels of fusion are also presented [2 4 6 12ndash17] Fusionof iris and fingerprint has attracted a lot of attention andresearchers have presented variety of approaches in the

literature [7 8 16 18 19] Baig et al [8] in 2009 proposed aframework formultimodal biometric fusion based on utiliza-tion of a single matcher implementation for both modalities(iris andfingerprint) For their experiment they used theWestVirginia Universityrsquos multimodal database containing 400images (4 enrolment images times 100 users) and the threshold isset to the equal error rate EERThe comparison is beingmadein terms of percentage improvement in EER rather than theEER values themselves Jagadeesan et al [16] in 2010 intro-duced a technique for cryptographic key generation by fusingfingerprint and iris biometrics The fingerprint extractor isminutia based while the iris extractor is based on canny edgedetector and Hough transform (Daugmanrsquos approach) Theminutiae points and texture properties were first extractedfrom fingerprint and iris images respectively and then theywere fused at the feature level to obtain the multibiometrictemplate and subsequently a 256-bit secure cryptographic keyfrom the multibiometric template is generated

In 2011 Jameer Basha et al [18] introduced a newframework for iris and fingerprint fusion at rank levelthey conducted experimental tests using three implementedfusion methods highest rank method Borda count methodand logistic regression method Their work achieved the bestexecution time required to match which is equal to 045seconds for the highest rank method with optimal FAR andFRR equal respectively to 0 and 025

In 2012 Radha and Kavitha [19] presented a novel fusionscheme of fingerprint and iris modalities at feature extractionlevel The scheme uses a concatenated feature vector fromboth iris and fingerprintThe logGabor filter is used to extractthe feature vectors of both modalities then the phase datafrom 1D log Gabor filters is extracted and quantized to fourlevels to encode the unique pattern of iris andFingerprint intobitwise biometric template Hamming distance (HD) is usedto generate a final match score Experimental results wereverified on database of 50 users accounting to FAR = 0 andFRR = 43The execution time required tomatch is reducedto 014 seconds

In 2013 Abdolahi et al [7] presented a multimodalbiometric system (fingerprint and iris) using fuzzy logic andweighted code After converting fingerprint and iris imagesto a binary code the decision level fusion is used to combinethe results Fingerprint code is weighed as 20 and iris codeas 80 The work achieved 2 FAR and FRR and 983accuracy

3 State of the Art of Multimodal Biometric

In this section we summarize the main ideas and principlesinvolved in the area of multimodal biometric recognition

31 Multimodal Biometrics versus Multibiometrics As ex-plained by most research papers in the field of biometricrecognition [5 12 16 20] the term ldquomultimodal biometricrdquorefers to multiple biometric traits used together at a specificlevel of fusion to recognize persons The ldquomultibiometricsrdquoincludes either the use of multiple algorithms also calledclassifiers at enrolment or matching stages for the samebiometric trait or the use of multiple sensors of the same


Multiplealgorithms

sensors

Repeatedinstances

Multiplemodalities

Scenariosof fusion










Sensorlevel

Featurelevel

Rank

Decision Score

level

levellevel

Fusion levels






(1)


(2)






Iris DB


Irisinput

EnhancementFeature

extractionmodule

Matchingmodule

Iris score


Fusionchoice


fusionmodule


Fingerprintinput



Fingerprint

DB

Featureextraction

module

Matchingmodule













Fusion ofscores

Threshold

Comparison

Result






HD =sum119873

119895=1119883119895 (XOR) 119884119895 (AND) 119883119899

1015840

119895(AND) 1198841198991015840

119895

119873 minus sum119873

119896=1119883119899119896 (OR) 119884119899119896

(3)


119895and 119884


119883119899119895and119884119899


119895and 119884

119895 and119873 is the


ED = radic119899

sum

119894=1

(119883119894minus 119884119894)

2

(4)

where119883119894and 119884



1198781015840=

119899

sum

119896=1

119878 (5)



1198781015840= 1205721198781+ (1 minus 120572) 1198782

(6)






For that








Result


decision

Decisionsfusion

Fuzzyinference

Fuzzy if then rules


Deg

ree o

f mem

bers

hip

Good Medium Bad1

0

Threshold

S1 S2 S3 S4 S5 S6







































































































100

80

60

40

20

0

1 12 14 16 18 19

FAR ()FRR ()





100

90

80

70

60

50

40

30

20

10

0

05 075 08 085 09 1

FAR ()FRR ()






(7)




Experiment Accuracy





Author andreference



Featureextraction



FAR = 0FRR = 512

Match time = 356 s




distance)

FAR = 0FRR = 43

Match time = 014 s








FAR = 0FRR = 005



8 Conclusion
















References







































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in




Multiplealgorithms

sensors

Repeatedinstances

Multiplemodalities

Scenariosof fusion










Sensorlevel

Featurelevel

Rank

Decision Score

level

levellevel

Fusion levels






(1)


(2)






Iris DB


Irisinput

EnhancementFeature

extractionmodule

Matchingmodule

Iris score


Fusionchoice


fusionmodule


Fingerprintinput



Fingerprint

DB

Featureextraction

module

Matchingmodule













Fusion ofscores

Threshold

Comparison

Result






HD =sum119873

119895=1119883119895 (XOR) 119884119895 (AND) 119883119899

1015840

119895(AND) 1198841198991015840

119895

119873 minus sum119873

119896=1119883119899119896 (OR) 119884119899119896

(3)


119895and 119884


119883119899119895and119884119899


119895and 119884

119895 and119873 is the


ED = radic119899

sum

119894=1

(119883119894minus 119884119894)

2

(4)

where119883119894and 119884



1198781015840=

119899

sum

119896=1

119878 (5)



1198781015840= 1205721198781+ (1 minus 120572) 1198782

(6)






For that








Result


decision

Decisionsfusion

Fuzzyinference

Fuzzy if then rules


Deg

ree o

f mem

bers

hip

Good Medium Bad1

0

Threshold

S1 S2 S3 S4 S5 S6







































































































100

80

60

40

20

0

1 12 14 16 18 19

FAR ()FRR ()





100

90

80

70

60

50

40

30

20

10

0

05 075 08 085 09 1

FAR ()FRR ()






(7)




Experiment Accuracy





Author andreference



Featureextraction



FAR = 0FRR = 512

Match time = 356 s




distance)

FAR = 0FRR = 43

Match time = 014 s








FAR = 0FRR = 005



8 Conclusion
















References







































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in




Iris DB


Irisinput

EnhancementFeature

extractionmodule

Matchingmodule

Iris score


Fusionchoice


fusionmodule


Fingerprintinput



Fingerprint

DB

Featureextraction

module

Matchingmodule













Fusion ofscores

Threshold

Comparison

Result






HD =sum119873

119895=1119883119895 (XOR) 119884119895 (AND) 119883119899

1015840

119895(AND) 1198841198991015840

119895

119873 minus sum119873

119896=1119883119899119896 (OR) 119884119899119896

(3)


119895and 119884


119883119899119895and119884119899


119895and 119884

119895 and119873 is the


ED = radic119899

sum

119894=1

(119883119894minus 119884119894)

2

(4)

where119883119894and 119884



1198781015840=

119899

sum

119896=1

119878 (5)



1198781015840= 1205721198781+ (1 minus 120572) 1198782

(6)






For that








Result


decision

Decisionsfusion

Fuzzyinference

Fuzzy if then rules


Deg

ree o

f mem

bers

hip

Good Medium Bad1

0

Threshold

S1 S2 S3 S4 S5 S6







































































































100

80

60

40

20

0

1 12 14 16 18 19

FAR ()FRR ()





100

90

80

70

60

50

40

30

20

10

0

05 075 08 085 09 1

FAR ()FRR ()






(7)




Experiment Accuracy





Author andreference



Featureextraction



FAR = 0FRR = 512

Match time = 356 s




distance)

FAR = 0FRR = 43

Match time = 014 s








FAR = 0FRR = 005



8 Conclusion
















References







































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in






HD =sum119873

119895=1119883119895 (XOR) 119884119895 (AND) 119883119899

1015840

119895(AND) 1198841198991015840

119895

119873 minus sum119873

119896=1119883119899119896 (OR) 119884119899119896

(3)


119895and 119884


119883119899119895and119884119899


119895and 119884

119895 and119873 is the


ED = radic119899

sum

119894=1

(119883119894minus 119884119894)

2

(4)

where119883119894and 119884



1198781015840=

119899

sum

119896=1

119878 (5)



1198781015840= 1205721198781+ (1 minus 120572) 1198782

(6)






For that








Result


decision

Decisionsfusion

Fuzzyinference

Fuzzy if then rules


Deg

ree o

f mem

bers

hip

Good Medium Bad1

0

Threshold

S1 S2 S3 S4 S5 S6







































































































100

80

60

40

20

0

1 12 14 16 18 19

FAR ()FRR ()





100

90

80

70

60

50

40

30

20

10

0

05 075 08 085 09 1

FAR ()FRR ()






(7)




Experiment Accuracy





Author andreference



Featureextraction



FAR = 0FRR = 512

Match time = 356 s




distance)

FAR = 0FRR = 43

Match time = 014 s








FAR = 0FRR = 005



8 Conclusion
















References







































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in





























































































100

80

60

40

20

0

1 12 14 16 18 19

FAR ()FRR ()





100

90

80

70

60

50

40

30

20

10

0

05 075 08 085 09 1

FAR ()FRR ()






(7)




Experiment Accuracy





Author andreference



Featureextraction



FAR = 0FRR = 512

Match time = 356 s




distance)

FAR = 0FRR = 43

Match time = 014 s








FAR = 0FRR = 005



8 Conclusion
















References







































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in





Experiment Accuracy





Author andreference



Featureextraction



FAR = 0FRR = 512

Match time = 356 s




distance)

FAR = 0FRR = 43

Match time = 014 s








FAR = 0FRR = 005



8 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 Comparative Study of Multimodal Biometric...

Documents

Transcript of Research Article Comparative Study of Multimodal Biometric...