Research Article Identifying Individuals Using ... - Hindawi

Hindawi Publishing CorporationJournal of EngineeringVolume 2013 Article ID 539284 8 pageshttpdxdoiorg1011552013539284

Research ArticleIdentifying Individuals Using Eigenbeat Features ofElectrocardiogram

Yogendra Narain Singh1 and Sanjay Kumar Singh2

1 Department of Computer Science amp Engineering Institute of Engineering amp Technology Gautam Buddh Technical UniversityLucknow 226 021 India

2Department of Computer Engineering Indian Institute of Technology Banaras Hindu University IIT (BHU) Varanasi 225 021 India

Correspondence should be addressed to Yogendra Narain Singh singhyngmailcom

Received 15 August 2012 Revised 12 February 2013 Accepted 12 February 2013

Academic Editor Karim Kabalan

Copyright copy 2013 Y N Singh and S K Singh This is an open access article distributed under the Creative Commons AttributionLicense which permits unrestricted use distribution and reproduction in any medium provided the original work is properlycited

The authors of this paper present a new method to characterize the electrocardiogram (ECG) for individual identification Wepropose an ECG biometric system which is insensitive to noise signals and muscle flexure The method utilizes the principal oflinearly projecting the heartbeat features into a subspace of lower dimension using an orthogonal basis that represents the mostsignificant features to distinguish the individuals The performance of the proposed biometric system is evaluated on the subjectsof both health statuses such as the ECG recordings of MIT-BIH Arrhythmia database and the ECG recordings of normal subjectsprepared at IIT(BHU) The result demonstrates that the derived eigenbeat features from proposed ECG characterization performbetter and achieve the recognition accuracy of 9142 and 9555 on the subjects of MIT-BIH Arrhythmia database and IIT(BHU)database respectively

1 Introduction

The accurate and automatic authentication of individualsis becoming inevitable in several aspects of our daily lifeIt includes border crossing business or commercial trans-actions health care physical access control and managingthe digital rights The proliferation of computers Internetand computer-based applications such as e-business e-commerce and e-learning are raised the concerns of securitybreaches and identity theft crimes Conventional methodsof automatic identity proofing use the credentials suchas passwords and PIN numbers The deployment of suchidentity proofing systems that are either possession basedor knowledge based raise the serious risk of identity theftIn addition the use of multiple documents (eg passportPAN card license ration card etc) for identity proofingoffers an opportunity to someone replicating one of theseidentity markers and pretending to be you The cases ofidentity theft are increased significantly which is a dark sideof documentary-proofing methods for identity recognition

Identity theft is a growing problem globally In a reportof Federal Trade Commission [1] identity theft made upabout one-fifth of all the consumer complaints reported in2010 According to the report the most popular type ofidentity theft involved criminals stealing victimrsquos identity sothat they could use their information to apply for federalbenefits such as social security payments The Federal TradeCommission estimated that about 3ndash5 of US residents havetheir identities stolen every year and surprisingly most ofthemmight not aware that this has happened to themOveralllosses from identity fraud in theUS alone are estimated to $37billion [2]There is no authentic data available to estimate theidentity fraud losses in countries like India or Brazil whereincreasing adoption of Internet and computer applicationsare grown exponentially in an insecure manner

With the recent advancement in technology it is nowpossible to create an automatic individual recognition sys-tem using biometric attributes such as face fingerprintiris or handwritten signatures that can be verified on-line[3] Individual authentication using biometrics is attractive

2 Journal of Engineering

because the authentication process is principally based onphysiological or behavioral characteristics that are uniqueand measurable Biometrics is being emerged as a state ofthe art tool of information security for accurate and efficientidentification of individuals in a digital society [4] Biometricidentities are intrinsic to an individual therefore these aredifficult to share and distribute among peers steal and forgeto the fraudulents counterfeit and hack [5]They are not easyto fool nor they are intrusive But biometric authenticationas a part of security is not a solution in itself

Biometric attributes are unique among individuals butthey are not the secret Biometric information is irrevocableand hard to regain identity [6] The biometric technology isbeing popular but it has the growing concerns of vulnerabili-ties that breach the security of the system and user privacyIt includes circumvention replay attacks and obfuscation[7] Circumvention refers to the reproduction of falsifiedcredentials from an original biometric sample whereas thepresentation of original biometric feature from an illegitimatesubject is referred to a replay attackThe removal of biometricfeatures in order to avoid the establishment of true identity iscalled obfuscation

This paper advocates to use the bioelectrical signal suchas electrocardiogram (ECG) as a novel biometric attributefor secured identity proofing The main advantage of usingan ECG as a biometrics is the robustness to circumventionreplay and obfuscation attacks If we succeed in establishingthe ECG signal as a biometrics then the recognition systemusing the ECG can be empowered with an inherent shieldto the security threats The ECG is a physiological signalgenerated in human heart that has an inherent feature ofvitality which signifies the life signs The ECG signal as abiometrics is sufficiently nonvulnerable to spoof attacks so itmay insure robustness of a system It is universally presentamong live subjects and as such it is naturally secured TheECG is difficult to mimic and hard to be copied or stolenTherefore the ECG has strong credentials to successfullyaddress the security and privacy issues of an individual [8]

The ECG acquired from different individuals show het-erogeneous characteristics [9ndash15]Theheterogeneity has beenmarked in the studies conducted for diagnosing arrhythmiapresent in the heart function [16] The distinctiveness of theECG signal is generally resulted due to the change in ionicpotential time of ionic potential to spread from differentpart of the heart muscles plasma levels of electrolytes (egpotassium calcium and magnesium etc) and the rhythmicdifferences The difference in the heart structure such aschest geometry position size and physical condition amongindividuals also manifest the distinctiveness in their heart-beats rhythm The distinct characteristics of an individualheartbeat are reflected in the change in morphology differ-ence in amplitudes and the variation in time intervals ofthe dominant fiducials The main issue to use the ECG asa biometrics is the variations present in the heartbeats andaccumulation of signal with noise artifacts making the datarepresentation more difficult [17 18]

In this paper we present a novel method for identifyingindividuals using their ECG signals in particular the methodis insensitive to noise and muscle flexure that are usually

contaminated to them We make use of ECG waveformdelineators [19 20] for determining the dominant fiducialsfrom each heartbeat efficiently The ECG characterization isperformed next such that the heartbeat interval features andthe morphological features are derived from the successivebeats In order to make the features insensitive to noise andother artifacts a two-stage procedure is employed In thefirst stage the information of amplitude features are derivedfrom the signal which is scaled using Pareto normalization[21] whereas in the second stage a linear projection ofderived heartbeat features from high-dimensional space to alower dimensional feature space is derived It uses principalcomponent analysis (PCA) also known as Karhunen-Loevemethods [22] for dimensionality reduction and yields projec-tion directions thatmaximize the scatter across all traces of anindividual ECG When the heartbeat features are projectedinto the subspace spanned by the dominant eigenvectorsthe separability among the subjects are manifested Thedimension of generated eigenvectors is the same as originalfeature vectors therefore they can be referred to as eigenbeatfeatures

The identity classification is performed using a nearestneighbor classifier The classification results are obtainedusing a subset of the MIT-BIH Arrhythmia database [23]and the ECG database prepared from normal subjects atIIT(BHU) In the sections to follow the dependency of recog-nition performance on the number of principal componentsis reported in both the databases We found extremely betterresults at the lower dimension of feature subspace of the ECGsignal

The rest of the paper is organized as follows The relatedwork and the state-of-the-art using the ECG signal asa biometrics is given in Section 2 Section 3 presents themethod of ECG characterization for biometric applicationThe schematic description of an ECG biometric recognitionsystem is presented in Section 4 The experimental resultsthat prove the efficacy of the proposed characterization of theECG signal for identity recognition on publically availabledatabase and the database acquired from real subjects arepresented in Section 5 Finally some conclusions are drawnin Section 6

2 Related Work

Different studies have shown that the ECG can be usedas a new candidate of biometrics for individual authenti-cation ([9ndash15]) Biel et al [9] amongst the first who havedemonstrated the use of ECG for biometric applicationTheyhave conducted the biometric experiment on a group of 20subjects and a multivariate method is used for classificationThe feasibility of using the ECG as a new biometrics forindividual identity verification has been shown by Shen etal [10] They have performed the experiment on appearanceand time domain features of the heartbeat However most ofthe features are extracted from QRS complex those are stablewith the change in the heart rateThe feature QT interval thatvaries with the heart rate is normalized Template matchingand decision-based neural network approaches are used to

Journal of Engineering 3

RR intervalST

segmentR

TP

QS

PRinterval

QTinterval

Pwave

QRScomplex

Twave

TPsegment

Figure 1 A typical ECG signal that includes three successive heartbeats and the information lying in the P Q R S and T waves

quantify the identity verification rate that are reported to be95 and 80 respectively After combining the classificationapproaches the result of identity verification is found to be100 for a group of 20 individuals

Israel et al [11] have shown that the ECG of an individualexhibits distinct pattern They have performed the ECGprocessing for quality check and a quantifiable metrics isproposed for classifying heartbeats among individuals Atotal of 15 intrabeat features based upon cardiac physiologyare extracted from each heartbeat and the classificationis performed using linear discriminant analysis The testsshow that the extracted features are independent to electrodepositions (eg around chest and neck) invariant to anindividualrsquos state of anxiety and unique to an individual

Wang et al [12] have introduced a two-step fiducial detec-tion framework that incorporates analytic and appearance-based features from the heartbeat The analytic featurescapture local information of a heartbeat that consist oftemporal and amplitude features while the appearance-basedfeatures capture the holistic patterns of a heartbeat To betterutilize the complementary characteristics of analytic andappearance based features a hierarchical data integrationscheme has been presented The method used for featureextraction is based on the combination of autocorrelation(AC) and discrete cosine transform (DCT) which is free fromfiducial detection The recognition performance of ACDCTmethod is reported between 9447 and 978

The feasibility of ECG signal to aid in human identi-fication has been explored by Singh and Gupta ([13ndash15])recently Signal processing methods are used to delineateECG waveforms (eg P and T waves) from each heartbeatThe delineation results are found optimum and stable incomparison to other published results These delineators areused along with QRS complex to extract different features ofclasses time interval amplitude and angle from clinicallydominant fiducials on each heartbeat They have conductedthe experiment on 50 subjectsrsquo ECG recordings of Physionetdatabase [23] The individuals are classified with an accuracyup to 99

3 Method of ECG Characterization

The ECG is a noninvasive tool used to record the electricalmanifestation of the contractile and relaxation activity of theheart Nobel laureate Willem Einthoven was the first whohad recorded the ECG in 1903 [24]The ECG can be recordedwith the surface electrodes placed on the limbs and the chestThe ECG devices use a varying number of electrodes rangingfrom 3 to 12 for signal acquisition while systems using moreelectrodes exceeding 12 and up to 120 are also available [25]Each normal cycle of an ECG signal contains P QRS and Twaves (eg see Figure 1) The P wave is a representation ofcontraction of the atrial muscle and has a duration of 60ndash100milliseconds (ms) It has low amplitude morphology of 01ndash025 millivolts (mV) and is usually found in the beginning ofthe heartbeatTheQRS complex is the result of depolarizationof themessy ventricles It is a sharp biphasic or triphasic waveof 80ndash120ms duration and shows a significant amplitudedeflection that varies from person to person The time takenfor an ionic potential to spread from sinus node throughthe atrial muscles and entering the ventricles is 120ndash200msand known as PR interval The ventricles have a relativelylong ionic potential duration of 300ndash420ms known as theQT interval The plateau part of ionic potential of 80ndash120msafter the QRS and known as the ST segment The return ofthe ventricular muscle to its resting ionic state causes the Twave that has an amplitude of 01ndash05mV and duration of120ndash180ms The duration from resting of ventricles to thebeginning of the next cycle of atrial contraction is knownas the TP segment which is a long plateau part of negligibleelevation

31 Feature Extraction Prior to use the ECG signal in thesubsequent stage of processing of heartbeat segmentation andfeatures extraction all signals are passed through a two-stagemedian filters of width 200ms and 600ms respectively toremove the baselinewanderThefirstmedian filter suppressesthe QRS complexes and P waves while the second medianfilter suppresses the T waves The resulting signal is then


subtracted from the original signal to produce the baselinecorrected ECG signal [26]

The QRS complex delineator is used to detect theheartbeats from the ECG signal We employ the techniqueproposed by Pan and Tompkins [27] for QRS complexdetection with some improvements It uses digital analysisof slope amplitude and width information of the ECGwaveforms The beginning and the end of the QRS complexthat is QRSonset and QRSoffset time instances respectivelyare delineated according to the location and convexity ofthe R peak Once the heartbeats are detected temporal timewindows are defined heuristically before and after the QRScomplex time instances to seek for the P and T waves Thetechnique proposed in [19] is used to determine the Ponset andPoffset time instances from the P wave while the techniqueproposed in [20] is used to determine Tonset and Toffset timeinstances from theTwaveThrough all these time instances ofthe heartbeats three different classes of features are derivedThese are (1) heartbeat interval features (2) interbeat intervalfeatures and (3) ECG morphological features(1)Heartbeat Interval Features Five features relating to heart-beat intervals are computed after heartbeat segmentationThe QRS width is the duration between the QRSonset andthe QRSoffset The T wave duration is defined as the timeinterval between theQRSoffset and theToffsetThePQ segmentis defined as the time interval between the Ponset and theQRSonset The pre-TP segment is defined as the time intervalbetween a given Ponset and the previous wave Toffset Similarlythe post-TP segment is defined as the time interval betweena given Toffset and the following wave Ponset(2) Interbeat Interval Features Ten features relating to inter-heartbeat intervals are computed after the segmentationof successive heartbeat fiducials points These features areextracted from the PP QQ SS TT and RR sequence of thesuccessive heartbeats The pre-PP (post-PP) interval is thetime interval between Ponset of a given heartbeat and the Ponsetof the previous (following) heartbeatThe pre-QQ (post-QQ)interval is the time interval betweenQPeak of a given heartbeatand the Qpeak of the previous (following) heartbeat The pre-SS (post-SS) interval is the time interval between Speak ofa given heartbeat and the Speak of the previous (following)heartbeat The pre-TT (post-TT) offset interval is the timeinterval between Toffset of a given heartbeat and the Toffsetof the previous (following) heartbeat Similarly the pre-RR(post-RR) interval is defined as the RR interval between agiven heartbeat and the previous (following) heartbeat(3)ECGMorphological FeaturesWe divide the ECGmorpho-logical features into two groups where both groups containthe amplitude values of the segmented heartbeat of the ECGsignal The first group contains thirty-three features Thesefeatures are determined within the time windows as shownin Figure 2 The first window is set between the QRSonset andthe QRSoffset Five features are extracted corresponding tothe fiducials of QRSonset Qpeak Rpeak 119878peak and QRSoffsetThe boundaries of the second window is set such that itapproximately covers the P wave It contains the portion ofthe heartbeat between the Ponset and the Ponset+120msUsing

+ 120 ms

FP

Toffset

QRSoffsetQRSonset

PonsetPonset

Figure 2 Extraction of ECG morphological features from a heart-beat where the fiducial point (FP) represents the position of R peak

FPminus80ms +100ms

+150ms FP +FP minus 420 ms240 ms

Figure 3 Extraction of ECGmorphological features from the scaledsamples of a heartbeat

linear interpolation method thirteen features are estimateduniformly within the time window Similarly the thirdwindow is bounded by the QRSoffset and the Toffset Fifteenfeatures of the heartbeat amplitude is derived uniformlywithin the window using linear interpolation

The second group contains twenty-eight features whichare extracted from the normalized ECG signal In the nor-malized signal the amplitude difference from 119909

119899119879to themean

120583 is measured in units of standard deviation 120590 such as

119909119899119879

1015840

=119909119899119879minus 120583

radic120590 (1)

where 119909119899119879

represents the data sample of size 119899 at discreteinstance of time 119879 [21]The aim of normalization is to reducethe sensitivity of the ECG signal both to noise and muscleflexure that are contaminated in the signal We define threedifferent time windows with respect to the location of theheartbeat fiducial points (FP) as shown in Figure 3 The firstwindow approximately covers the QRS complex and it iscontained the portion of the ECG signal between FP minus 80msand FP + 100ms A total of nine features are derived fromthe ECG signalThe secondwindow approximately covers theP wave and it is extended to FP minus 240ms from FP minus 80mstowards left Again a total of nine features are resulted withinthe window The third window approximately contains the T


ECG delineationECG signal PreprocessingHeartbeat

Segmentation

P and T wavesdelineation

Feature extractionselectionInterval features

Morphological features

Authentication

Database templateversus

probe sample

Decisionmaking

Qualitycheck

Eigenbeatfeatures

Figure 4 Schematic of a biometric recognition system for identifying individuals based on their electrocardiograms

wave and it is started from FP + 150ms to FP + 420ms Tenamplitude features are derived from this window In all threewindows the features are derived using linear interpolationmethod where the signal is sampled uniformly

32 Selection of Eigenbeat Features The eigenbeat method isbased on the linear projection of the sample space to a lowdimensional feature space [22] It uses principal componentanalysis (PCA) for dimensionality reduction that yields theprojection direction that maximizes the scatter across allsamples present in the gallery and the probe ECG signals

More formally let us consider that there be 119873 classes offeature vectors 119864

1 1198642 119864

119873 where each class 119864

119903contains

one or more feature vectors (119903 = 1 2 119873) in an 119899-dimensional space Then a set of 119898 (119898 lt 119899) feature basisvectors 120601119898

119897=1can be estimated by maximizing the expression

argmax120601

10038161003816100381610038161003816120601119879

11987812060110038161003816100381610038161003816 (2)

where 120601119897| 119897 = 1 2 119898 is the set of 119899-dimensional

eigenvectors of the scatter matrix 119878 corresponding to the 119898largest eigenvalues where 119878 is defined as

119878 =1

119873

119873

sum119903=1

(119864119903minus 120583) (119864

119903minus 120583)119879

(3)

where 120583 (isin R119899) = 1119873sum119873

119903=1119864119903is the mean of all feature

vectors participated in the recognition process It is to benoted that the dimension of the generated eigenvectors is thesame as the original feature vectors therefore they can bereferred to as eigenbeat features The generated eigenvectorsform the basis representation of the gallery and the probeECG signals They yield projection directions that maximizethe scatter across all feature vectors within a subject Thecoefficients set 119862

119903are derived for each class of feature

vectors corresponding to each subject 119903 which is a compactrepresentation of the heartbeat features in the gallery setIf a class contains more than one feature vector then theaverage of all heartbeats for a single subject provides thegallery representation 119862

119903against which the probe data is to

be compared For identity recognition the classification isperformed using a nearest neighbor classifier in the reducedfeature space The best match in the gallery set is the choiceof subject 119903 that minimizes the distance between 119862

119903and 119862

119901

such thatarg min

119894

10038171003817100381710038171003817119862119903minus 119862119901

10038171003817100381710038171003817 (4)

where 119862119901is a vector of coefficients in eigenspace for probe

ECG signal which can be obtained using similar processingsteps as used by the gallery ECG

4 ECG Biometric Recognition System

The biometric recognition system of identifying individualsusing the ECG signal is shown in Figure 4 ECG signalsacquired from the individuals are preprocessed for qualitycheck It makes necessary correction of the signal from noiseand muscle flexure The ECG delineation includes segmen-tation of heartbeats such as detection of the P Q R S andT waves and determination of their end fiducials The featureextraction includes determination of the interval features andthe ECG morphological features from the successive beatsand derived the eigenbeat features Finally the authenticationis performed on reduced feature set in the projected domaincomparing the features of the gallery and the probe ECGsignals using nearest neighbor criterion

5 Experimental Results

The performance of the aforementioned identity recognitionsystem is tested on two different databases The first databaseis prepared from publically available PhysioBank archives[23] in particular MIT-BIH Arrhythmia database is usedForty-four ECG recordings are randomly selected from thisdatabase in this studyThe second database is prepared in thelaboratory of the School of Biomedical Engineering IndianInstitute of Technology (Banaras Hindu University) usingthe PowerLab 425 system of AD Instruments A total 29volunteers aged 20 to 56 years are participated in the dataenrollment process and the data are acquired in multiplesessions across a year The data acquisition is performed ina more simplistic manner with the subjects merely sittingon a chair or a wooden stool under relaxed condition andthe clamp electrodes are fixed to both wrists and left ankleThe data are bandpass filtered at 03ndash50Hz and sampled at1000Hz

The MIT-BIH Arrhythmia database contains only oneECG recording for each subjects therefore the completerecord of a subject is divided into two halves such that thefirst half is used for training and latter half is used for testingIn IIT(BHU) database two different sessions of data areused for the gallery and the probe We randomly select 10sets of heartbeats from the gallery data and the features arederived from the successive occurrences of 10 beats such thatthey meet the delineators requirement in the each set Priorto apply the selection procedure of eigenbeat features thederived features are normalized using Z-score criterion [28]A representation relative to the basis formed by dominanteigenvectors is derived by selecting the five most significanteigenvectors corresponding to five maximum eigenvalues


S1S2S3

S4S5

0minus2000

0

800

PC1

PC2

(a)

M1M2M3

M4M5

0

minus200

0

1000PC1

PC2

(b)

Figure 5 Intersubject variability represented by first and second principal components of five different subjects (a) MIT-BIH Arrhythmiadatabase and (b) IIT(BHU) database

885

899

913

927

941

955

1 2 3 4 5Number of principal components

Reco

gniti

on ac

cura

cy (

)

IIT(BHU) databaseMIT-BIH Arrhythmia database

Figure 6 Subjects ECG recognition performance of MIT-BIH Arrhythmia database and IIT(BHU) database Variations in performance ofeigenbeat features depend on the number of principal components

Finally the coefficients of the components in the projectedsubspace are generated which is a compact representation ofheartbeats in the gallery set Averaging over all of the heart-beat sets for a single subject provide the gallery representationagainst which the probe data is to be compared The probesignal undergoes the same processing steps as the galleryset and derives a representation relative to the basis formedby the dominant eigenvectors The best match in the galleryset is the choice of the subject that minimizes the distance(Euclidean) between components

The distinction between eigenbeat features among thesubjects of MIT-BIH Arrhythmia database and IIT(BHU)database can be represented by principal components that are

shown in Figure 5 The separability between the subjects ofboth health statuses is clearly visible at the lower dimensionsof projection which are represented by considering only firstprincipal component (PC1) and second principal component(PC2) For example a decrease in intraclass variability andan increase in interclass separability represented by principalcomponents such as PC1 and PC2 for the subjects of MIT-BIHArrhythmia database and IIT(BHU) database are shownin Figures 5(a) and 5(b) respectively

The performance of the proposed system on eigenbeatfeatures of MIT-BIH Arrhythmia database and IIT(BHU)database is shown in Figure 6 through a plot of recogni-tion accuracy versus the number of principal components


The recognition accuracy can be computed as the inverse ofan equal error rate reported by the system For the subjectsof MIT-BIH Arrhythmia database the system is reported arecognition accuracy of 885when the comparison betweenthe gallery and the probe ECG is done on the informationderived by first principal component (PC1) The recognitionaccuracy can be improved further if the system accumulatesinformation associated to other principal components Forexample the reported values of recognition accuracy are8987 9078 9123 and 9142 for PC2 PC3 PC4 andPC5 respectively A similar trend is observed for the subjectsof IIT(BHU) database The recognition accuracy is reportedmaximum to 9555 for the subjects of IIT(BHU) databaseon the accumulation of first five principal components (PC5)whereas the minimum accuracy of 9115 is reported at PC1The recognition accuracy at other dimension of principalcomponents such as PC2 PC3 and PC4 are found to be9365 9482 and 9522 respectively The reason beingthat of getting higher recognition accuracy for the subjectsof IIT(BHU) database is that it contains the ECG data ofhealthy subjects that are acquired under normal conditionsThis confirms that the proposed characterization of ECGsignal and subsequently derived eigenbeat features are robustenough to distinguish the subjects of both health statusessuch as the healthy subjects or the subjects suffering formcardiac arrhythmia

6 Conclusion

This study has proposed a new method to characterize theECG signal for identifying individuals The set of featureshave derived from the analysis of successive heartbeats whichinclude the heartbeat interval features and the waveformmorphological features We have derived eigenbeat featuresusing the method of linearly projecting the sample space toa lower dimensional feature space The advantages of usingeigenbeat features are the elimination of noise and muscleflexure from the ECG data reducing the complexity to accessa larger attribute set and simplifying the classification processThe reported results have proved the effectiveness of pro-posed characterization of the ECG signal and subsequentlyderived eigenbeat features for individual identification

References

[1] Federal Commission report February 2012 httpwwwftcgovsentinelreportssentinel-annual-reportssentinelcy2009pdf

[2] ldquoA report by Javelin Strategy amp Researchrdquo Washington PostFebruary 2011

[3] Y N Singh ldquoChallenges of UID Environmentrdquo in Proceedingsof the UID National Conference on Impact of Aadhaar inGovernance Computer Society of India pp 37ndash45 LucknowIndia December 2011

[4] Y N Singh and S K Singh ldquoThe state of information securityrdquoin Proceedings of the Artificial Intelligence and Agents Theoryand Applications (AIATA rsquo11) pp 363ndash367 Varanasi IndiaDecember 2011

[5] A K Jain A Ross and S Pankanti ldquoBiometrics a toolfor information securityrdquo IEEE Transactions on InformationForensics and Security vol 1 no 2 pp 125ndash143 2006

[6] A Watson ldquoBiometrics easy to steal hard to regain identityrdquoNature vol 449 no 7162 p 535 2007

[7] Y N Singh and S K Singh ldquoA taxonomy of biometricsystem vulnerabilities and defensesrdquo International Journal ofBiometrics vol 5 no 2 pp 137ndash159 2013

[8] Y N Singh S K Singh and P Gupta ldquoFusion of electrocar-diogram with unobtrusive biometrics an efficient individualauthentication systemrdquo Pattern Recognition Letters vol 33 no14 pp 1932ndash1941 2012

[9] L Biel O Pettersson L Philipson and P Wide ldquoECG analysisa new approach in human identificationrdquo IEEE Transactions onInstrumentation and Measurement vol 50 no 3 pp 808ndash8122001

[10] T W Shen W J Tompkins and Y H Hu ldquoOne-lead ECG foridentity verificationrdquo in Proceedings of the 2nd Joint EngineeringinMedicine andBiology 24thAnnual Conference and theAnnualFall Meeting of the Biomedical Engineering Society Conference(EMBSBMES rsquo02) pp 62ndash63 Houston Tex USA October2002

[11] S A Israel J M Irvine A Cheng M DWiederhold and B KWiederhold ldquoECG to identify individualsrdquo Pattern Recognitionvol 38 no 1 pp 133ndash142 2005

[12] Y Wang F Agrafioti D Hatzinakos and K N PlataniotisldquoAnalysis of human electrocardiogram for biometric recogni-tionrdquo EURASIP Journal on Advances in Signal Processing vol2008 Article ID 148658 pp 1ndash11 2008

[13] Y N Singh and P Gupta ldquoBiometric method for human iden-tification using electrocardiogramrdquo in Advances in Biometricsvol 5558 of Lecture Notes of Computer Science pp 1270ndash12792009

[14] Y N Singh and P Gupta ldquoECG to individual identificationrdquoin Proceedings of the 2nd IEEE International Conference onBiometricsTheory Applications and Systems (BTAS rsquo08) pp 1ndash8October 2008

[15] Y N Singh and P Gupta ldquoCorrelation-based classification ofheartbeats for individual identificationrdquo Soft Computing vol 15no 3 pp 449ndash460 2011

[16] J R HamptonThe ECGMade Easy Churchill Livingstone 5thedition 2001

[17] Y N Singh and S K Singh ldquoBioelectrical signals as emergingbiometrics issues and challengesrdquo ISRN Signal Processing vol2012 Article ID 712032 13 pages 2012

[18] Y N Singh and S K Singh ldquoEvaluation of electrocardiogramfor biometric authenticationrdquo Journal of Information Securityvol 3 no 1 pp 39ndash48 2012

[19] Y N Singh and P Gupta ldquoA robust delineation approachof electrocardiographic P wavesrdquo in Proceedings of the IEEESymposium on Industrial Electronics and Applications (ISIEArsquo09) vol 2 pp 846ndash849 October 2009

[20] Y N Singh and P Gupta ldquoAn efficient and robust technique ofT wave delineation in electrocardiogramrdquo in Proceedings of the2nd International Conference on Bio-Inspired Systems and SignalProcessing (BIOSIGNALS rsquo09) pp 146ndash154 January 2009

[21] R A van den Berg H C J Hoefsloot J A Westerhuis AK Smilde and M J van der Werf ldquoCentering scaling andtransformations improving the biological information contentof metabolomics datardquo BMC Genomics vol 7 article 142 pp 1ndash15 2006


[22] R O Duda P E Hart and D G Stork Pattern ClassificationWiley Delhi India 2nd edition 2009

[23] Physionet Physiobank archives Massachusetts Institute ofTechnology Cambridge January 2011 httpwwwphysionetorgphysiobankdatabaseecg

[24] L A Geddes and R A Roeder ldquoThe first electronic electrocar-diographrdquo Cardiovascular Engineering vol 2 no 2 pp 73ndash792002

[25] P Zarychta F E Smith S T King et al ldquoBody surfacepotential mapping for detection of myocardial infarct sitesrdquo inProceedings of the IEEE Computers in Cardiology (CAR rsquo07) vol34 pp 181ndash184 October 2007

[26] P De Chazal M OrsquoDwyer and R B Reilly ldquoAutomatic clas-sification of heartbeats using ECG morphology and heartbeatinterval featuresrdquo IEEE Transactions on Biomedical Engineeringvol 51 no 7 pp 1196ndash1206 2004

[27] J Pan and W J Tompkins ldquoA real-time QRS detection algo-rithmrdquo IEEE Transactions on Biomedical Engineering vol 32no 3 pp 230ndash236 1985

[28] Y N Singh and P Gupta ldquoQuantitative evaluation of normal-ization techniques of matching scores in multimodal biometricsystemsrdquo in Advances in Biometrics vol 4642 of Lecture Notesof Computer Science pp 574ndash583 2007

International Journal of

AerospaceEngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

RoboticsJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014


Active and Passive Electronic Components

Control Scienceand Engineering

Journal of



RotatingMachinery


Hindawi Publishing Corporation httpwwwhindawicom

Journal ofEngineeringVolume 2014

Submit your manuscripts athttpwwwhindawicom

VLSI Design



Shock and Vibration


Civil EngineeringAdvances in

Acoustics and VibrationAdvances in



Electrical and Computer Engineering

Journal of

Advances inOptoElectronics


Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

SensorsJournal of


Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014


Chemical EngineeringInternational Journal of Antennas and

Propagation




Navigation and Observation



DistributedSensor Networks



because the authentication process is principally based onphysiological or behavioral characteristics that are uniqueand measurable Biometrics is being emerged as a state ofthe art tool of information security for accurate and efficientidentification of individuals in a digital society [4] Biometricidentities are intrinsic to an individual therefore these aredifficult to share and distribute among peers steal and forgeto the fraudulents counterfeit and hack [5]They are not easyto fool nor they are intrusive But biometric authenticationas a part of security is not a solution in itself

Biometric attributes are unique among individuals butthey are not the secret Biometric information is irrevocableand hard to regain identity [6] The biometric technology isbeing popular but it has the growing concerns of vulnerabili-ties that breach the security of the system and user privacyIt includes circumvention replay attacks and obfuscation[7] Circumvention refers to the reproduction of falsifiedcredentials from an original biometric sample whereas thepresentation of original biometric feature from an illegitimatesubject is referred to a replay attackThe removal of biometricfeatures in order to avoid the establishment of true identity iscalled obfuscation

This paper advocates to use the bioelectrical signal suchas electrocardiogram (ECG) as a novel biometric attributefor secured identity proofing The main advantage of usingan ECG as a biometrics is the robustness to circumventionreplay and obfuscation attacks If we succeed in establishingthe ECG signal as a biometrics then the recognition systemusing the ECG can be empowered with an inherent shieldto the security threats The ECG is a physiological signalgenerated in human heart that has an inherent feature ofvitality which signifies the life signs The ECG signal as abiometrics is sufficiently nonvulnerable to spoof attacks so itmay insure robustness of a system It is universally presentamong live subjects and as such it is naturally secured TheECG is difficult to mimic and hard to be copied or stolenTherefore the ECG has strong credentials to successfullyaddress the security and privacy issues of an individual [8]

The ECG acquired from different individuals show het-erogeneous characteristics [9ndash15]Theheterogeneity has beenmarked in the studies conducted for diagnosing arrhythmiapresent in the heart function [16] The distinctiveness of theECG signal is generally resulted due to the change in ionicpotential time of ionic potential to spread from differentpart of the heart muscles plasma levels of electrolytes (egpotassium calcium and magnesium etc) and the rhythmicdifferences The difference in the heart structure such aschest geometry position size and physical condition amongindividuals also manifest the distinctiveness in their heart-beats rhythm The distinct characteristics of an individualheartbeat are reflected in the change in morphology differ-ence in amplitudes and the variation in time intervals ofthe dominant fiducials The main issue to use the ECG asa biometrics is the variations present in the heartbeats andaccumulation of signal with noise artifacts making the datarepresentation more difficult [17 18]

In this paper we present a novel method for identifyingindividuals using their ECG signals in particular the methodis insensitive to noise and muscle flexure that are usually

contaminated to them We make use of ECG waveformdelineators [19 20] for determining the dominant fiducialsfrom each heartbeat efficiently The ECG characterization isperformed next such that the heartbeat interval features andthe morphological features are derived from the successivebeats In order to make the features insensitive to noise andother artifacts a two-stage procedure is employed In thefirst stage the information of amplitude features are derivedfrom the signal which is scaled using Pareto normalization[21] whereas in the second stage a linear projection ofderived heartbeat features from high-dimensional space to alower dimensional feature space is derived It uses principalcomponent analysis (PCA) also known as Karhunen-Loevemethods [22] for dimensionality reduction and yields projec-tion directions thatmaximize the scatter across all traces of anindividual ECG When the heartbeat features are projectedinto the subspace spanned by the dominant eigenvectorsthe separability among the subjects are manifested Thedimension of generated eigenvectors is the same as originalfeature vectors therefore they can be referred to as eigenbeatfeatures

The identity classification is performed using a nearestneighbor classifier The classification results are obtainedusing a subset of the MIT-BIH Arrhythmia database [23]and the ECG database prepared from normal subjects atIIT(BHU) In the sections to follow the dependency of recog-nition performance on the number of principal componentsis reported in both the databases We found extremely betterresults at the lower dimension of feature subspace of the ECGsignal

The rest of the paper is organized as follows The relatedwork and the state-of-the-art using the ECG signal asa biometrics is given in Section 2 Section 3 presents themethod of ECG characterization for biometric applicationThe schematic description of an ECG biometric recognitionsystem is presented in Section 4 The experimental resultsthat prove the efficacy of the proposed characterization of theECG signal for identity recognition on publically availabledatabase and the database acquired from real subjects arepresented in Section 5 Finally some conclusions are drawnin Section 6

2 Related Work

Different studies have shown that the ECG can be usedas a new candidate of biometrics for individual authenti-cation ([9ndash15]) Biel et al [9] amongst the first who havedemonstrated the use of ECG for biometric applicationTheyhave conducted the biometric experiment on a group of 20subjects and a multivariate method is used for classificationThe feasibility of using the ECG as a new biometrics forindividual identity verification has been shown by Shen etal [10] They have performed the experiment on appearanceand time domain features of the heartbeat However most ofthe features are extracted from QRS complex those are stablewith the change in the heart rateThe feature QT interval thatvaries with the heart rate is normalized Template matchingand decision-based neural network approaches are used to


RR intervalST

segmentR

TP

QS

PRinterval

QTinterval

Pwave

QRScomplex

Twave

TPsegment












+ 120 ms

FP

Toffset

QRSoffsetQRSonset

PonsetPonset


FPminus80ms +100ms







119909119899119879

1015840

=119909119899119879minus 120583

radic120590 (1)

where 119909119899119879




Segmentation




Authentication


probe sample

Decisionmaking

Qualitycheck

Eigenbeatfeatures





1 1198642 119864


119903contains



argmax120601

10038161003816100381610038161003816120601119879

11987812060110038161003816100381610038161003816 (2)



119878 =1

119873

119873

sum119903=1

(119864119903minus 120583) (119864

119903minus 120583)119879

(3)

where 120583 (isin R119899) = 1119873sum119873







119903and 119862

119901

such thatarg min

119894

10038171003817100381710038171003817119862119903minus 119862119901

10038171003817100381710038171003817 (4)









S1S2S3

S4S5

0minus2000

0

800

PC1

PC2

(a)

M1M2M3

M4M5

0

minus200

0

1000PC1

PC2

(b)


885

899

913

927

941

955


Reco

gniti

on ac

cura

cy (

)









6 Conclusion


References
































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










RR intervalST

segmentR

TP

QS

PRinterval

QTinterval

Pwave

QRScomplex

Twave

TPsegment












+ 120 ms

FP

Toffset

QRSoffsetQRSonset

PonsetPonset


FPminus80ms +100ms







119909119899119879

1015840

=119909119899119879minus 120583

radic120590 (1)

where 119909119899119879




Segmentation




Authentication


probe sample

Decisionmaking

Qualitycheck

Eigenbeatfeatures





1 1198642 119864


119903contains



argmax120601

10038161003816100381610038161003816120601119879

11987812060110038161003816100381610038161003816 (2)



119878 =1

119873

119873

sum119903=1

(119864119903minus 120583) (119864

119903minus 120583)119879

(3)

where 120583 (isin R119899) = 1119873sum119873







119903and 119862

119901

such thatarg min

119894

10038171003817100381710038171003817119862119903minus 119862119901

10038171003817100381710038171003817 (4)









S1S2S3

S4S5

0minus2000

0

800

PC1

PC2

(a)

M1M2M3

M4M5

0

minus200

0

1000PC1

PC2

(b)


885

899

913

927

941

955


Reco

gniti

on ac

cura

cy (

)









6 Conclusion


References
































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation












+ 120 ms

FP

Toffset

QRSoffsetQRSonset

PonsetPonset


FPminus80ms +100ms







119909119899119879

1015840

=119909119899119879minus 120583

radic120590 (1)

where 119909119899119879




Segmentation




Authentication


probe sample

Decisionmaking

Qualitycheck

Eigenbeatfeatures





1 1198642 119864


119903contains



argmax120601

10038161003816100381610038161003816120601119879

11987812060110038161003816100381610038161003816 (2)



119878 =1

119873

119873

sum119903=1

(119864119903minus 120583) (119864

119903minus 120583)119879

(3)

where 120583 (isin R119899) = 1119873sum119873







119903and 119862

119901

such thatarg min

119894

10038171003817100381710038171003817119862119903minus 119862119901

10038171003817100381710038171003817 (4)









S1S2S3

S4S5

0minus2000

0

800

PC1

PC2

(a)

M1M2M3

M4M5

0

minus200

0

1000PC1

PC2

(b)


885

899

913

927

941

955


Reco

gniti

on ac

cura

cy (

)









6 Conclusion


References
































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation











Segmentation




Authentication


probe sample

Decisionmaking

Qualitycheck

Eigenbeatfeatures





1 1198642 119864


119903contains



argmax120601

10038161003816100381610038161003816120601119879

11987812060110038161003816100381610038161003816 (2)



119878 =1

119873

119873

sum119903=1

(119864119903minus 120583) (119864

119903minus 120583)119879

(3)

where 120583 (isin R119899) = 1119873sum119873







119903and 119862

119901

such thatarg min

119894

10038171003817100381710038171003817119862119903minus 119862119901

10038171003817100381710038171003817 (4)









S1S2S3

S4S5

0minus2000

0

800

PC1

PC2

(a)

M1M2M3

M4M5

0

minus200

0

1000PC1

PC2

(b)


885

899

913

927

941

955


Reco

gniti

on ac

cura

cy (

)









6 Conclusion


References
































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










S1S2S3

S4S5

0minus2000

0

800

PC1

PC2

(a)

M1M2M3

M4M5

0

minus200

0

1000PC1

PC2

(b)


885

899

913

927

941

955


Reco

gniti

on ac

cura

cy (

)









6 Conclusion


References
































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation











6 Conclusion


References
































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation



















RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation









Research Article Identifying Individuals Using ... - Hindawi

Documents

Transcript of Research Article Identifying Individuals Using ... - Hindawi