ECG Signals for Biometric Applications

Are we there yet?

Carlos Carreiras

, Andr

e Lourenc¸o

1,2

, Ana Fred

and Rui Ferreira

Instituto de Telecomunicac¸

oes, Av. Rovisco Pais 1, Lisboa, Portugal

Instituto Superior de Engenharia de Lisboa, R. Cons. Em

ıdio Navarro 1, Lisboa, Portugal

Hospital de Santa Marta, R. de Santa Marta 50, Lisboa, Portugal

Keywords:

Biometrics, Person Recognition, ECG, Classiﬁcation.

Abstract:

The potential of the electrocardiographic (ECG) signal as a biometric trait has been ascertained in the literature

over the past decade. The inherent characteristics of the ECG make it an interesting biometric modality,

given its universality, intrinsic aliveness detection, continuous availability, and inbuilt hidden nature. These

properties enable the development of novel applications, where non-intrusive and continuous authentication

are critical factors. Examples include, among others, electronic trading platforms, the gaming industry, and

the auto industry, in particular for car sharing programs and ﬂeet management solutions. However, there are

still some challenges to overcome in order to make the ECG a widely accepted biometric. In particular, the

questions of uniqueness (inter-subject variability) and permanence over time (intra-subject variability) are still

largely unanswered. In this paper we focus on the uniqueness question, presenting a preliminary study of our

biometric recognition system, testing it on a database encompassing 618 subjects. We also performed tests

with subsets of this population. The results reinforce that the ECG is a viable trait for biometrics, having

obtained an Equal Error Rate of 9.01% and an Error of Identiﬁcation of 15.64% for the entire test population.

1 INTRODUCTION

Over the past decade, numerous groups have demon-

strated the potential of electrocardiographic (ECG)

signals for identity recognition applications (Biel

et al., 2001; Kyoso and Uchiyama, 2001; Silva et al.,

2007a; Wang et al., 2008). Due to its inherent char-

acteristics, the ECG signal is emerging as an interest-

ing biometric trait, given that, following the proper-

ties deﬁned in (Jain et al., 1999), it can be found in

all living humans (Universality), it has been shown

to perform accurately for subsets of the population

(Performance), and it can be easily obtained using ap-

propriate devices (Measurability). These sensors can

be designed in a non-intrusive way (Acceptability),

in particular when using an Off-the-Person approach

(Silva et al., 2013a). Furthermore, the ECG is not eas-

ily spoofed (Circumvention), as it does not depend on

any external body traits, provides intrinsic aliveness

detection, and is continuously available.

These properties of the ECG signal enable the

development of novel and interesting applications,

where non-intrusive and continuous authentication

are critical factors. Examples of such applications

include electronic trading platforms, where high-

security, continuous authentication is essential, in the

gaming industry, where the ECG sensor could be in-

tegrated into the game controller itself to identify the

players in a multi-player scenario, and in the auto in-

dustry, particularly for car sharing programs and ﬂeet

management solutions.

At the moment, the biggest challenges faced by

the ECG as a biometric trait relate to its Permanence

and Uniqueness, and the question remains if this

modality is ready for real-world applications. While

Permanence deals with the question of temporal in-

variance of the templates, that is, intra-subject vari-

ability, Uniqueness pertains to the discernibility of

the templates from different subjects, that is, inter-

subject variability. Studies on the permanence ques-

tion can already be found in the literature, for instance

in (Agraﬁoti et al., 2011; Silva et al., 2013c). In this

paper, we present a preliminary study on the unique-

ness question. We accomplish this by testing our

recognition system on an ECG signal database with

618 subjects, the biggest to date to be used for bio-

metrics, to the best of our knowledge. We also per-

formed tests with subsets of this population, assessing

765

Carreiras C., Lourenço A., Fred A. and Ferreira R..

ECG Signals for Biometric Applications - Are we there yet?.

DOI: 10.5220/0005160507650772

In Proceedings of the 11th International Conference on Informatics in Control, Automation and Robotics (IVC&ITS-2014), pages 765-772

ISBN: 978-989-758-040-6

 2014 SCITEPRESS (Science and Technology Publications, Lda.)

the behavior of the recognition system with a varying

number of subjects.

The remainder of this paper is organized as fol-

lows: Section 2 provides an overview of the charac-

teristics of the ECG signal and its use in biometric

systems; Section 3 describes the methodology used

for the biometric recognition system, including a de-

scription of the database used, feature extraction, and

classiﬁcation approaches; Section 4 summarizes the

obtained experimental results; and Section 5 outlines

the main conclusions.

2 BACKGROUND

It is widely known that the basic function of the heart

is to pump blood throughout the body, demanding a

highly synchronized sequence of muscular contrac-

tions. These are initiated by small electrical currents

that propagate through the myocardium’s cells, orig-

inating an electrical signal that can be recorded at

the body surface (the ECG). These potentials can be

measured by placing two electrodes on the body’s

surface, determining the voltage difference between

them (Webster, 2009). Different electrode placements

produce different perspectives of the heart, termed

leads or derivations, given the spatial characteristics

of the heart’s electrical ﬁeld and how it propagates

throughout the body (Neuman, 1998).

The ECG is a semi-periodic signal, with each cy-

cle being characterized by the typical P-QRS-T heart-

beat waveform. The signal as a whole has a rich in-

formation content, being a wellbeing and health in-

dicator, and is related with the psychophysiological

state (Carreiras et al., 2013b). In order to have a co-

herent clinical diagnostic tool, the lead placement has

been standardized. Much of the standard system is

based on Einthoven’s groundbreaking work, with the

use of the three limb leads (I, II, and III), as the limbs

are easily identiﬁed anatomical references (Webster,

2009). Additionally, the augmented leads (aVF, aVL,

aVR) and the six precordial leads (V1 – V6) are also

typically recorded in clinical settings.

In the context of ECG biometrics, current ap-

proaches found in the literature can be classiﬁed as

either ﬁducial or non-ﬁducial (Agraﬁoti et al., 2011;

Odinaka et al., 2012; Silva et al., 2013b). The for-

mer describes methods based on reference points in

the signals (Israel et al., 2005; Shen et al., 2002; Silva

et al., 2007b; Oliveira and Fred, 2009), while the lat-

ter methods rely on intrinsic information within the

ECG signal, without having any particular cues as ref-

erence (Chan et al., 2008; Chiu et al., 2008; Wang

et al., 2008; Coutinho et al., 2013). Partially ﬁdu-

(a) ECG trace

(b) Heartbeat waveform templates

Figure 1: Examples of (a) of a ECG trace (Subject A), and

(b) the heartbeat waveform templates for two distinct indi-

viduals.

cial methods, like our approach presented below, rely

on ﬁducial information only for ECG segmentation

(Wang et al., 2008; Lourenc¸o et al., 2011; Carreiras

et al., 2013a; Silva et al., 2013c). We refer the reader

to (Agraﬁoti et al., 2011; Odinaka et al., 2012; Silva

et al., 2013b) and references therein for a compre-

hensive literature review. Figure 1 shows an example

of an ECG signal trace, and the segmented heartbeat

templates for two distinct subjects, where the differ-

ences between them are apparent.

One signiﬁcant contribution to the usefulness and

acceptability of the ECG as a biometric trait is the use

of an Off-the-Person approach for signal acquisition

(Silva et al., 2013a). In this approach, only one ECG

lead is used, with the signal being acquired at the hand

palms or ﬁngers, using just two (non-gelled) contact

points, as opposed to multiple contact points through-

out the body using gelled electrodes. The lead place-

ment in this case is non-standard, however it has been

shown to be highly correlated with the standard lead

I (Carreiras et al., 2013c). Various research groups

have used this approach (Chan et al., 2008; Lourenc¸o

et al., 2011). However, the signals obtained with this

setup are harder to analyze, as they are more suscep-

tible to noise artifacts due to unstable electrode-skin

ICINCO2014-11thInternationalConferenceonInformaticsinControl,AutomationandRobotics

766

Table 1: Five largest ECG biometrics studies found in (Odinaka et al., 2012), with the reported authentication (AP) and

identiﬁcation (IP) performance; AUR: Area Under ROC curve; EER: Equal Error Rate; NA: Not Available.

Study Sample Size ECG Lead AP (%) IP (%)

(Zhang and Wei,

2006)

502

I NA 85.3

II NA 92.0

V1 NA 95.2

V2 NA 97.4

(Odinaka et al.,

2010)

269 Electrodes placed

on lower ribcage

0.37 (EER) 99

(Shen et al.,

2010)

168 I (hands) NA 95.3

(Saﬁe et al.,

2011)

112 I 94.54 (AUR) NA

(Irvine et al.,

2003)

104 NA NA 91

contact and electromyographic (EMG) activity.

The Off-the-Person approach enables the seamless

integration of the ECG sensor into everyday objects.

One such example, as shown in Figure 2, is the in-

tegration of the ECG sensor into the steering wheel

of a car using conductive textiles. In this car shar-

ing demonstrator, the user, in order to authenticate on

the system, touches the contactless member card on a

reader to provide the assumed identity. This identity

is then validated through the ECG signal by simply

placing the hands on the steering wheel, as in a nor-

mal driving situation. On successful authentication,

various user-speciﬁc conﬁgurations could be loaded,

such as preferred radio stations, mirror positions, and

address lists, among others.

Figure 2: Integration of an Off-the-person ECG sensor into

the steering wheel of a car; the electrodes are highlighted in

red.

Regarding the uniqueness problem, there are cur-

rently no studies assessing the performance of ECG

biometric systems encompassing very large data sets,

such as the work done for iris recognition by Daug-

man, encompassing more than 600 000 different iris

patterns (Daugman, 2006). Using the review by Odi-

naka et al. (Odinaka et al., 2012) as source, ECG bio-

metric studies use, on average, databases of about 50

subjects. Table 1 provides a list of the ﬁve largest

studies, with the reported authentication and identiﬁ-

cation performances. Unfortunately, for various rea-

sons mainly related to privacy concerns, many stud-

ies use in-house databases, which are not publicly

available. Additionally, most public ECG databases,

notably the ones available on Physionet (Goldberger

et al., e 13), were built for research on pathophysiol-

ogy, not biometrics, with most of the records having

some kind of heart pathology. For this reason, as is

described in the next section, we were forced to ob-

tain our own database, in order to test our recognition

methodology with a larger number of subjects.

3 METHODOLOGY

3.1 Database

Our research group entered into a collaboration with a

local hospital (Hospital de Santa Marta) specialized in

cardiac issues, with the goal of obtaining a large ECG

database. The records thus obtained are acquired dur-

ing normal hospital operation, encompassing sched-

uled appointments, emergency cases, and bedridden

patients. Therefore, most of the records represent

pathological cases.

The signals were acquired using Philips

PageWriter Trim III devices, following the tra-

ditional 12 lead placement, with a sampling rate

of 500 Hz and 16bit resolution. Each record has a

duration of 10 seconds. To date, we have received,

over a period of 10 months, 4 332 records from 2 055

distinct subjects, whose true identities are obfuscated

at the hospital.

ECGSignalsforBiometricApplications-Arewethereyet?

767

(a) Rate of normal vs pathological subjects and records

(b) Gender distribution

Figure 3: Population statistics of the database, for a total of

4 332 records and 2 055 subjects, with (a) the rate of normal

subjects and records, subject (b) gender, and (c) age distri-

butions; the whiskers in the boxplots extend to the lowest

and highest data points still within 1.5 times the interquar-

tile range.

As a ﬁrst step, for this paper we decided to fo-

cus only on the healthy individuals. Consequently,

each record had to be labeled by a specialist either as

normal or pathological. Of all the records, 832 were

deemed normal, corresponding to 618 subjects. This

ﬁgure surpasses by about 100 the largest ECG biomet-

rics study currently found in the literature. Figure 3

summarizes the relevant population statistics.

Note that, although our target applications follow

the Off-the-Person approach, such a large database

takes a lot of time and effort to obtain, requiring clear-

ance by an ethics committee, ﬁnding volunteers, sign-

ing of informed consent forms, among others. Nev-

ertheless, if we cannot demonstrate the potential, in

regards to the uniqueness question, of the ECG as a

biometric in higher quality signals, then certainly that

is not possible with hand signals.

3.2 ECG Biometric System

The typical block diagram of a ﬁducial, or partially

ﬁducial, biometric system is depicted in Figure 4.

These systems rely on the detection of notable ECG

complexes for segmentation and extraction of a se-

quence of individual heartbeats. Typically, the QRS

complex is used for that purpose.

1-lead

ECG reader

Filtering

QRS

Detection

Classifier

Decision

Outlier

Removal

Pattern

Extraction

Enrollment Stage

Recognition Stage

Template

Storage

ECG

Acquisition

Figure 4: Block diagram of a typical ECG biometric sys-

tem.

Our ECG biometric system, designed with hand

ECG signals in mind, starts with the acquisition of

raw data, in this case the lead I ECG signal. The

acquired signal is then submitted to a data prepro-

cessing block, which performs a digital ﬁltering step

(band pass FIR ﬁlter order 150, and cutoff frequencies

[5;20] Hz) and the QRS complex detection (Lourenc¸o

et al., 2012). The outputs of this block are segmented

individual heartbeats, and a RR interval time series.

Given that segmentation algorithms are not per-

fect, especially for noisy signals like the ones ob-

tained from the hands, we implement an outlier detec-

tion block, which performs detection and removal of

anomalous ECG heartbeats. We follow the DMEAN

approach described in (Lourenc¸o et al., 2013), which

ICINCO2014-11thInternationalConferenceonInformaticsinControl,AutomationandRobotics

768

computes the distance of all templates in a recording

session to the mean template for that session, with

templates being considered outliers if the computed

distance is higher than an adaptive threshold.

The pattern extraction block takes the prepro-

cessed input signals, and starts by aligning all the

heartbeat waveforms by their R-peak instants, and by

clipping them in the interval [−200; 400]ms around

that instant. In the scope of this work, we consider

the features to be all the amplitudes within this inter-

val.

Finally, a k-NN classiﬁer (with k=3) is used to-

gether with the cosine distance metric, to produce a

decision on the recognition of the individual (either in

authentication or identiﬁcation), as it was found to be

a good compromise between performance and com-

putational cost (Silva et al., 2012). Altogether, our

biometric system is fairly simple, being computation-

ally light and opening the possibility of integrating it

into embedded systems, which have limited process-

ing power.

4 RESULTS

We evaluated the performance of the biometric sys-

tem for both the identiﬁcation and authentication sce-

narios. For the identiﬁcation scenario, we computed

the Error of Identiﬁcation (EID), which corresponds

to the number of incorrect identiﬁcations normalized

by the total number of tests. For the authentication

scenario, we computed, for each operating point of

the classiﬁer (each distance threshold), the False Ac-

ceptance Rate (FAR), the False Rejection Rate (FRR),

the True Acceptance Rate (TAR), and the True Rejec-

tion Rate (T RR), given by

FAR =

T N+FP

, FRR =

T P+FN

TAR =

T P

T P+FP

, T RR =

T N

T N+FP

(1)

where T P and T N are the number of true positives

and negatives, and FP and FN are the number of

false positives and negatives, respectively. From these

rates, we estimate the Equal Error Rate (EER), which

corresponds to the operating point for which the FAR

is equal to the FRR, using piecewise polynomial in-

terpolation.

Furthermore, we used a leave-one-out (LOO) ap-

proach for cross validation (Efron, 1983), given the

fact that the number of templates for some subjects

was low (minimum of 4 templates), enabling us to

maximize the number of templates to train the clas-

siﬁer, which requires at least 3 templates (3-NN). In

order to to this, we selected a random group of 4

templates for each subject, which are then partitioned

with the LOO method. We repeated this procedure

10 times, computing the average authentication and

identiﬁcation performance across all runs.

Additionally, we assessed the behavior of the sys-

tem with subsets of the population, encompassing 5,

10, 20, 30, 40, and 50 subjects. These subsets cor-

respond to our targeted applications, ranging from a

small group (e.g. in a multiplayer game setting, or a

family sharing a car) to small businesses (e.g. a local

distribution company). The subjects in each subgroup

were randomly selected from the initial population,

repeating this process 150 times, each run following

the cross validation method described above.

The results obtained for the entire population

(P618) are presented in Table 2, comparing them to

a previous baseline experiment performed using a

smaller database (63 subjects), which uses signals ob-

tained at the hands, making obvious the costs in per-

formance resulting from the use of hand signals. Re-

garding the EID, the value obtained is on par with the

results presented in (Zhang and Wei, 2006) for lead I

signals (see Table 1), with the added bonus of using a

larger database.

Table 2: EER and EID obtained for the entire test popula-

tion (P618) and the baseline experiment (63 subjects, hand

ECG).

Case EER (%) EID (%)

P618 9.01 15.64

Baseline 13.26 36.40

Figure 5 shows the evolution of the FAR and FRR

with the authentication distance threshold, as well as

the Receiver Operating Characteristic (ROC) curve,

which plots the the TAR against the FAR, highlighting

an Area Under ROC (AUR) curve of 95.51%, similar

to the one obtained in (Saﬁe et al., 2011). Also of note

in Figure 5(a) is the fact that the FAR increases more

slowly than the FRR decreases with the threshold.

Results for the population subsets are presented in

Figure 6. Figure 6(a) highlights the fact that the EER

does not seem to be affected by the population size.

On the other hand, Figure 6(b) shows that the EID

increases with the increasing number of subjects.

5 CONCLUSIONS

Research to date has demonstrated that the ECG sig-

nal, due to its intrinsic nature, has the potential to

complement existing person recognition approaches

ECGSignalsforBiometricApplications-Arewethereyet?

769

(a) Authentication EER

(b) Autentication ROC curve

Figure 5: Authentication results for the entire population,

with (a) the EER determination, and (b) the ROC curve.

(a multibiometrics scenario), and, in some settings, to

be used as a single modality.

However, the ﬁeld is lacking a thorough exami-

nation of the limits of this modality in regards to the

number of subjects, that is, we need to know if the

information that we can extract from the ECG is suf-

ﬁcient to distinguish a large population. This paper is

a contribution to that goal, assessing the performance

of our ECG biometric system, which was designed for

an Off-the-Person sensor approach, in a database with

618 subjects, examining as well the effect of the pop-

ulation size on the performance of the system, using

subsets of the test population.

The results of our work indicate a performance of

our system on par with similar studies found in the lit-

erature, with an Equal Error Rate of 9.01% and an Er-

ror of Identiﬁcation of 15.64% for the entire test pop-

ulation. We also demonstrated that, while the authen-

tication performance does not degrade with increas-

ing number of subjects, the same does not happen

(a) Authentication EER

(b) Identiﬁcation EID

Figure 6: Results obtained for the population subsets in the

(a) authentication, and the (b) identiﬁcation scenarios.

with the identiﬁcation scenario, where the error pro-

gressively increases with increasing number of sub-

jects. Nevertheless, these results, together with the

latest developments in recognition methods, template

extraction and selection, and sensor devices, reinforce

that the ECG is a viable trait for biometric applica-

tions.

Our future work will focus on the study of sources

of intra-subject variability, in particular heart rate

changes and morphological shape alterations due to

pathological situations. Additionally, we will try to

improve the representativeness of the test population

in regards to age, and examine the performance of the

system when using the other standard ECG leads, ei-

ther independently or in combination (fusion of clas-

siﬁers).

ACKNOWLEDGEMENTS

This work was partially funded by Fundac¸

ao para a

ICINCO2014-11thInternationalConferenceonInformaticsinControl,AutomationandRobotics

770

encia e Tecnologia (FCT) under grants PTDC/EEI-

SII/2312/2012, SFRH/BD/65248/2009 and SFRH/PR

OTEC/49512/2009, whose support the authors grate-

fully acknowledge. We would also like to thank Joana

Santos for her hard work labeling the ECG records.

REFERENCES

Agraﬁoti, F., Gao, J., and Hatzinakos, D. (2011). Biomet-

rics, chapter Heart Biometrics: Theory, Methods and

Applications, Biometrics. InTech.

Biel, L., Petterson, O., Phillipson, L., and Wide, P. (2001).

ECG analysis: A new approach in human identiﬁca-

tion. IEEE Trans. Instrumentation and Measurement,

50(3):808–812.

Carreiras, C., Lourenc¸o, A., Silva, H., and Fred, A. L. N.

(2013a). A unifying approach to ECG biometric

recognition using the wavelet transform. In Int. Conf.

on Image Analysis and Recognition, pages 53–62.

Carreiras, C., Lourenc¸o, A., Aidos, H., Silva, H., and Fred,

A. (2013b). Morphological ecg analysis for attention

detection. In Int. Conf. on Neural Computation Theory

and Applications.

Carreiras, C., Lourenc¸o, A., Silva, H., and Fred, A.

(2013c). Comparative study of medical-grade and off-

the-person ecg systems. In Int. Congress on Cardio-

vascular Technologies.

Chan, A. D. C., Hamdy, M. M., Badre, A., and Badee, V.

(2008). Wavelet distance measure for person identiﬁ-

cation using electrocardiograms. IEEE Trans. on In-

strum. and Meas., 57(2):248–253.

Chiu, C.-C., Chuang, C.-M., and Hsu, C.-Y. (2008). A

novel personal identity veriﬁcation approach using a

discrete wavelet transform of the ECG signal. In Proc.

Int. Conf. on Multimedia and Ubiquitous Engineering,

pages 201–206. IEEE Computer Society.

Coutinho, D., Silva, H., Gamboa, H., Fred, A., and

Figueiredo, M. (2013). Novel ﬁducial and non-

ﬁducial approaches to electrocardiogram-based bio-

metric systems. Biometrics, IET, 2(2):64–75.

Daugman, J. (2006). Probing the uniqueness and random-

ness of iriscodes: Results from 200 billion iris pair

comparisons. Proc. of the IEEE, 94(11):1927–1935.

Efron, B. (1983). Estimating the error rate of a prediction

rule: Improvement on cross-validation. J. of the Amer-

ican Statistical Association, 78(382):316–331.

Goldberger, A. L., Amaral, L. A. N., Glass, L., Hausdorff,

J. M., Ivanov, P. C., Mark, R. G., Mietus, J. E., Moody,

G. B., Peng, C.-K., and Stanley, H. E. (2000 (June

13)). PhysioBank, PhysioToolkit, and PhysioNet:

Components of a new research resource for complex

physiologic signals. Circulation, 101(23):e215–e220.

Irvine, J., Israel, S., Wiederhold, M., and Wiederhold, B.

(2003). A new biometric: human identiﬁcation from

circulatory function. In Joint Statistical Meetings of

the American Statistical Association, San Francisco.

Israel, S., Irvine, J., Cheng, A., Wiederhold, M., and

Wiederhold, B. (2005). ECG to identify individuals.

Pattern Recognition, 38(1):133–142.

Jain, A. K., Bolle, R., and Pankanti, S. (1999). Biometrics :

personal identiﬁcation in networked society. Kluwer,

1st edition.

Kyoso, M. and Uchiyama, A. (2001). Development of an

ECG identiﬁcation system. In Proc. of the 23rd An-

nual Int. Conf. of the IEEE Engineering in Medicine

and Biology Society, volume 4, pages 3721–3723.

Lourenc¸o, A., Silva, H., Carreiras, C. C., and Fred, A. L. N.

(2013). Outlier detection in non-intrusive ECG bio-

metric system. In Int. Conf. on Image Analysis and

Recognition, pages 43–52.

Lourenc¸o, A., Silva, H., and Fred, A. (2011). Unveiling

the biometric potential of Finger-Based ECG signals.

Computational Intelligence and Neuroscience.

Lourenc¸o, A., Silva, H., Leite, P., Lourenc¸o, R., and Fred,

A. (2012). Real time electrocardiogram segmentation

for ﬁnger based ECG biometric. In Proc. of the Int.

Conf. on Bio-inspired Signals and Signal Processing

(BIOSIGNALS), pages 49–54.

Neuman, M. R. (1998). Biopotential ampliﬁers. Medical

instrumentation: application and design, pages 316–

318.

Odinaka, I., Lai, P.-H., Kaplan, A., O’Sullivan, J., Sirevaag,

E., Kristjansson, S., Shefﬁeld, A., and Rohrbaugh,

J. W. (2010). ECG biometrics: A robust short-time

frequency analysis. In Proc. of the IEEE Int. Work-

shop on Information Forensics and Security (WIFS),

pages 1–6.

Odinaka, I., Lai, P.-H., Kaplan, A., O’Sullivan, J., Sirevaag,

E., and Rohrbaugh, J. (2012). ECG biometric recog-

nition: A comparative analysis. IEEE Trans. on Infor-

mation Forensics and Security, 7(6):1812–1824.

Oliveira, C. and Fred, A. L. N. (2009). ECG-based authen-

tication: Bayesian vs. nearest neighbour classiﬁers. In

Proc. of the Int. Conf. on Bio-inspired Systems and

Signal Processing (BIOSIGNALS), pages 163–168.

Saﬁe, S. I., Soraghan, J. J., and Petropoulakis, L. (2011).

Electrocardiogram (ECG) biometric authentication

using pulse active ratio (par). IEEE Trans. on Infor-

mation Forensics and Security, 6(4):1315–1322.

Shen, T. W., Tompkins, W. J., and Hu, Y. H. (2002). One-

lead ECG for identity veriﬁcation. Proc. of the Int.

Conf. of the IEEE Engineering in Medicine and Biol-

ogy Society, 1:62–63.

Shen, T.-W. D., Tompkins, W. J., and Hu, Y. H. (2010). Im-

plementation of a one-lead ECG human identiﬁcation

system on a normal population. J. of Engineering and

Computer Innovations, 2(1):12–21.

Silva, H., Carreiras, C., Lourenc¸o, A., and Fred, A. L. N.

(2013a). Off-the-person electrocardiography. In

Int. Congress on Cardiovasular Technologies (CAR-

DIOTECHNIX), pages 99–106.

Silva, H., Gamboa, H., and Fred, A. (2007a). Applicability

of lead v

ECG measurements in biometrics. In Proc.

of the Int. eHealth, Telemedicine and Health ICT Fo-

rum (Med-e-Tel), pages 177–180.

ECGSignalsforBiometricApplications-Arewethereyet?

771

Silva, H., Gamboa, H., and Fred, A. (2007b). One lead

ECG based personal identiﬁcation with feature sub-

space ensembles. In Proc. of the 5th Int. Conf. on Ma-

chine Learning and Data Mining in Pattern Recogni-

tion, pages 770–783, Berlin. Springer.

Silva, H., Lourenc¸o, A., Canento, F., Fred, A., and Raposo,

N. (2013b). ECG biometrics: Principles and applica-

tions. In Proc. of the 6th Conf. on Bio-Inspired Sys-

tems and Signal Processing (BIOSIGNALS).

Silva, H., Lourenc¸o, A., and Fred, A. (2012). In-vehicle

driver recognition based on hand ECG signals. In

Proc. of the 2012 ACM Int. Conf. on Intelligent User

Interfaces, IUI ’12, New York, NY, USA. ACM.

Silva, H. P., Fred, A., Lourenco, A., and Jain, A. K.

(2013c). Finger ECG signal for user authentication:

Usability and performance. In 2013 IEEE Sixth Int.

Conf. on Biometrics: Theory, Applications and Sys-

tems (BTAS), pages 1–8. IEEE.

Wang, Y., Agraﬁoti, F., Hatzinakos, D., and Plataniotis,

K. N. (2008). Analysis of human electrocardiogram

for biometric recognition. EURASIP J. Adv. Signal

Process, 2008.

Webster, J. G. (2009). Medical Instrumentation Application

and Design. Wiley, 4th edition.

Zhang, Z. and Wei, D. (2006). A new ECG identiﬁcation

method using bayes’ teorem. In TENCON 2006. 2006

IEEE Region 10 Conference, pages 1–4. IEEE.

ICINCO2014-11thInternationalConferenceonInformaticsinControl,AutomationandRobotics

772