Compressed Domain ECG Biometric Identiﬁcation using JPEG2000

Yi-Ting Wu, Hung-Tsai Wu and Wen-Whei Chang

Institute of Communications Engineering, National Chiao-Tung University, Hsinchu, Taiwan

Keywords:

ECG Biometric, Person Identiﬁcation, JPEG2000.

Abstract:

In wireless telecardiology applications, electrocardiogram (ECG) signals are often represented in compressed

format for efﬁcient transmission and storage purposes. Incorporation of compressed ECG based biometric

enables faster person identiﬁcation as it by-passes the full decompression. This study presents a new method

to combine ECG biometrics with data compression within a common JPEG2000 framework. To this end, ECG

signal is considered as an image and the JPEG2000 standard is applied for data compression. Features relat-

ing to ECG morphology and heartbeat intervals are computed directly from the compressed ECG. Different

classiﬁcation approaches are used for person identiﬁcation. Experiments on standard ECG databases demon-

strate the validity of the proposed system for biometric identiﬁcation with high accuracies on both healthy and

diseased subjects.

1 INTRODUCTION

The demand for improved security for person identi-

ﬁcation has been growing rapidly and among the po-

tential alternatives is employing innovative biomet-

ric traits. Biometric identiﬁcation is reliant on pat-

tern recognition approaches by extracting physiolog-

ical or behavioral traits of human body and match-

ing them with enrollment database. Various biomet-

rics have been proposed for use in person identiﬁca-

tion, such as voice, face, and ﬁngerprint. However,

these biometrics either cannot provide reliable perfor-

mance or not robust enough against falsiﬁcation. Re-

cent studies have suggested the possibility of using

electrocardiogram (ECG) as a new biometric modal-

ity for person identiﬁcation (Biel et al., 2001; Israel

et al., 2005; Odinaka et al., 2012; Plataniotis et al.,

2006; Chiu et al., 2009). ECG signal is a record-

ing of the electrical activity of human heart, which

is individual-speciﬁc in the sense of amplitude and

time durations of the ﬁducial points. Furthermore,

ECG signal can deliver the proof of subject’s being

alive as extra information which other biometrics can-

not deliver as easily. It is believed that ECG bio-

metric would be particularly effective in health care

applications, as the signal can be used for diagno-

sis of cardiac diseases and also be used to identify

subjects before granting them medical services. Re-

cently, ECG biometric recognition has been success-

fully commercialized as products in mobile applica-

tions such as health care and online payment. For ex-

ample, Nymi wristband (Nymi Inc., 2013) is a wear-

able biometric authentication device that recognizes

unique ECG patterns and interfaces directly with mo-

bile devices as a replacement for passwords. In 2015,

Linear Dimensions also announced a family of bio-

metric authentication devices including ECG Biolock

and ECG optical wireless mouse (Linear Dimensions

Inc., 2015). Both devices offer proven security and

will authenticate users by learning their unique bio-

metric signature in ECG waveform pattern. Based on

the features that are extracted from ECG signals, we

can classify ECG biometrics as either ﬁducial points

dependent (Biel et al., 2001; Israel et al., 2005; Odi-

naka et al., 2012) or independent (Plataniotis et al.,

2006; Chiu et al., 2009). Fiducial-based approaches

rely on local features linked to the peak and time dura-

tions of the P-QRS-T waves. On the other hand, non-

ﬁducial approaches extract statistical features based

on the overall morphology of ECG waveform.

In wireless telecardiology scenarios, compressed

ECG packets are often preferred for efﬁcient trans-

mission and storage purposes. Most of ECG com-

pression methods adopt one-dimensional (1-D) repre-

sentation for ECG signals and focus on the utilization

of the intra-beat correlation between adjacent samples

(Jalaleddine et al., 1990). To better exploit both intra-

beat and inter-beat correlations, 2-D compression al-

gorithms have been proposed by converting ECG sig-

nals into data arrays and then applying vector quan-

tization (Sun and Tai, 2005) or the JPEG2000 im-

age coding standard (Bilgin et al., 2003). Irrespective

Wu Y., Wu H. and Chang W..

Compressed Domain ECG Biometric Identiﬁcation using JPEG2000.

DOI: 10.5220/0005499500050013

In Proceedings of the 12th International Conference on Signal Processing and Multimedia Applications (SIGMAP-2015), pages 5-13

ISBN: 978-989-758-118-2

 2015 SCITEPRESS (Science and Technology Publications, Lda.)

of the underlying method used for data compression,

compressed ECG imposes a new challenge for per-

son identiﬁcation as most of existing algorithms have

implicitly considered that biometric features are ex-

tracted from raw ECG signals (Biel et al., 2001; Israel

et al., 2005; Odinaka et al., 2012; Plataniotis et al.,

2006; Chiu et al., 2009). Full decompression is then

required to convert compressed data into ECG signals

prior to feature extraction. This step is undesirable in

health care systems, as the hospital may have thou-

sands of enrolled patients and decompression of all

their ECG packets is an enormous work. Thus, there

has been a new focus on biometric techniques which

directly read the compressed ECG to obtain unique

features with good discrimination power. Apart from

its advantage of by-passing the full decompression,

reduced template size also enables faster biometric

matching compared to the non-compressed domain

approaches. In (Suﬁ and Khalil, 2011), the authors

proposed a clustering method for compressed-domain

ECG biometric using specially designed compression

algorithms. The method starts with the detection of

cardiac abnormality and only the normal compressed

ECG data are used for person identiﬁcation. It is ex-

pected that considering the ECG as images and then

applying the JPEG2000 will lead to further better and

generalized results. As the discrete wavelet transform

(DWT) is an embedded part of the JPEG2000, and

DWT itself is one of the best features for ECG bio-

metrics (Chiu et al., 2009), working in DWT domain

remains to be the most promising area for compressed

ECG based biometric.

Another problem which requires further investiga-

tion is to test the feasibility of ECG biometrics with

diseased patients in cardiac irregular conditions. Pre-

vious works have shown that ECG biometric prob-

lem for healthy persons can be satisfyingly solved

with high recognition accuracies, but a much lower

accuracy may be achieved for cardiovascular disease

(CVD) patients. This is mainly because that CVD

may cause irrecoverable damage to the heart and in-

curs different forms of distorted ECG morphologies.

Recently, there have been initial studies of ECG bio-

metrics for diseased patients with ECG irregularities.

In (Chiu et al., 2009), the authors proposed a DWT-

based algorithm and reported overall accuracies of

100% and 81% on 35 normal subjects and 10 arrhyth-

mia patients, respectively. In (Agraﬁoti and Hatzi-

nakos, 2009), the authors obtained 96.42% recogni-

tion rate using autocorrelation method when tested on

26 healthy subjects from two databases and 30 pa-

tients with atrial premature contraction and premature

ventricular contraction. Another recent study (Yang

et al., 2013) indicated that a normalization and inter-

Figure 1: Block diagram of the 2-D ECG compression

scheme (Bilgin et al., 2003).

polation algorithm can achieve 100% and 90.11% in

accuracies on 52 healthy subjects and 91 CVD pa-

tients, respectively. The main purpose of this study is

to extract features in a way that the intra-subject vari-

ability is minimized and the inter-subject variability

is maximized.

The rest of this paper is organized as follows. Sec-

tion 2 describes the ECG fundamentals and presents

a preprocessor which converts ECG signals to 2-D

images. Also included is a short overview of the

JPEG2000 encoding algorithm. Details of the al-

gorithms for the proposed compressed-domain ECG

biometric system are provided in Section 3. Section

4 presents the experimental results on standard ECG

databases for both healthy and diseased subjects. Fi-

nally, Section 5 gives our conclusions.

2 2-D ECG DATA COMPRESSION

2-D ECG data compression algorithms are designed

to exploit both intra-beat and inter-beat correlations

of ECG signals. To begin, we apply a preprocessor

which can be viewed as a cascade of two stages. In

the ﬁrst stage, the QRS complex in each heartbeat

is ﬁrstly detected for segmenting and aligning 1-D

ECG signals to 2-D images and in the second stage,

the constructed ECG images are compressed by the

JPEG2000 (Taubman and Marcellin, 2001). Figure 1

shows the block diagram of the 2-D ECG compres-

sion scheme described in (Bilgin et al., 2003).

2.1 Signal Preprocessing

ECG itself is 1-D in the time-domain, but can be

viewed as a 2-D signal in terms of its implicit period-

icity. ECG signals tend to exhibit considerable simi-

larity between adjacent heartbeats, along with short-

term correlation between adjacent samples. Thus,

by dividing ECG signals into segments with lengths

equal to the heartbeats, there should be a large cor-

relation between individual segments. Typical ECG

waveform of a heartbeat consists of a P wave, a QRS

complex, and a T wave (Thaler, 2012). The QRS

complex is the most characteristic wave in an ECG

SIGMAP2015-InternationalConferenceonSignalProcessingandMultimediaApplications

waveform and hence, its peak can be used to iden-

tify each heartbeat. To begin, ECG signals are band-

pass ﬁltered to remove various noises. Afterwards

we apply the Biomedical Signal Processing Toolbox

(Aboy et al., 2002) to detect the R peak of each QRS

complex. Accordingly, ECG signals are divided into

heartbeat segments and each segment is stored as one

row of a 2-D data array. Having constructed the data

array as such, the intra-beat correlation is in the hor-

izontal direction of the array and the inter-beat cor-

relation is in the vertical direction. Since the heart-

beat segments may have different lengths, each row

of the data array is period normalized to a ﬁxed length

of N

= 200 samples via cubic spline interpolation.

This choice was based on the observation that the av-

erage heartbeat length is about 0.8 second, which cor-

responds to 200 samples for a sampling frequency of

250 Hz. Note that the original heartbeat lengths were

represented with 9 bits and transmitted as side infor-

mation. Finally, we proceed to construct ECG im-

ages of dimension N

× 200 by gathering together N

rows of the data array and normalizing the amplitude

of each component to an integer ranging from 0 to

255. The constructed grayscale ECG images are then

ready to be further compressed by the JPEG2000 cod-

ing standard.

2.2 JPEG2000 Encoding Algorithm

The JPEG2000 coding standard supports lossy and

lossless compression of grayscale and color imagery.

Although the standard was originally developed for

still image compression (Taubman and Marcellin,

2001), its applicability for ECG compression has been

proposed in (Bilgin et al., 2003). As shown in Figure

2, the JPEG2000 encoding process consists of several

operations: preprocessing, 2-D DWT, quantization,

entropy coding and bit-stream organization. It begins

with a preprocessor which divides the source image

into disjoint rectangular regions called tiles. For each

tile, the DC level of image samples is shifted to zero

and color space transform is performed to de-correlate

the color information. The 2-D DWT can be viewed

as applying a 1-D DWT decomposition along the lines

then the columns to generate an approximation sub-

band and three detail subbands oriented horizontally,

vertically and diagonally. With respect to the lifting

realization of 1-D DWT (Daubechies and Sweldens,

1998), prediction and update steps are performed on

the input signal to obtain the detail and the approx-

imation signals. A multiresolution representation of

the input image over J decomposition levels is ob-

tained by recursively repeating these steps to the re-

sulting approximation coefﬁcients.

Figure 2: Fundamental building blocks of JPEG2000 en-

coder (Taubman and Marcellin, 2001).

With J-level wavelet decomposition, the image

will have a total of 3J + 1 subbands. For notational

convenience, the subbands S

are numbered from 1

to 3J + 1, with 1 and 3J + 1 corresponding to the

top-left and bottom-right subbands, respectively. Let

= {s

(m,n),1 ≤ m ≤ M

,1 ≤ n ≤ N

} represent the

j-th subband whose row and column dimensions are

denoted by M

and N

with j ∈ {1,2,...,3J +1}. For

each coefﬁcient s

(m,n) located at position (m,n), a

mid-tread uniform quantizer is applied to obtain an

index v

(m,n) as follows:

(m,n) = sign[s

(m,n)] ·



(m,n)|

∆



, (1)

where ∆

denotes the quantizer step size for the j-th

subband. A different quantizer is employed for the

coefﬁcients of each subband and therefore, a bit al-

location among the subbands is carried out in order

to meet a targeted coding rate ρ. The last step in

JPEG2000 encoding consists in entropy coding of the

quantizer indexes with two tier encoders. The tier-1

encoder employs a context modeling to cluster the bits

of quantizer indexes into groups with similar statistics

in order to improve the efﬁciency of the arithmetic

coder. In the tier-2 encoder, the output of the arith-

metic coder is collected into packets and a bit-stream

is generated according to a predeﬁned syntax.

3 COMPRESSED DOMAIN ECG

BIOMETRIC

Person identiﬁcation is essentially a pattern recogni-

tion problem consisted of two stages: feature extrac-

tion and classiﬁcation. Under the JPEG2000 frame-

work, the person identiﬁcation problem is analo-

gous to a content-based image retrieval (CBIR) prob-

lem. Concerning compressed-domain biometric tech-

niques, the JPEG2000 code-stream is subject to par-

tial decoding and then features relating to ECG mor-

phology are computed directly from the dequantized

wavelet coefﬁcients. In the classiﬁcation stage, the

query ECG of an unknown subject will be compared

with the enrollment database to ﬁnd a match. The

CompressedDomainECGBiometricIdentificationusingJPEG2000

Figure 3: The proposed ECG biometric system.

block diagram of the proposed ECG biometric system

is shown in Figure 3.

3.1 Feature Extraction in DWT Domain

Feature extraction is the ﬁrst step in applying ECG

biometrics to person identiﬁcation and one that con-

ditions all the subsequent steps of system implemen-

tation. For large image databases, color, shape and

texture features are considered the most important

content descriptors in CBIR problems. Due to the

grayscale nature of ECG images, we only focus on

the texture features that characterize smooth, coarse-

ness and regularity of the speciﬁc image. One ef-

fective tool for texture analysis is the DWT as it

provides good time and frequency localization abil-

ity. Its multi-resolution nature also allows the de-

composition of an ECG image into different scales,

each of which represents particular coarseness of the

signal. Furthermore, DWT coefﬁcients can be ob-

tained without involving a full decompression of the

JPEG2000 code-stream. This is a favorable property

as the inverse DWT and subsequent decoding pro-

cesses could impose intensive computational burden.

Different texture features such as energy, signiﬁcance

map, and modelling of wavelet coefﬁcients at the out-

put of wavelet ﬁlter-banks have been successfully ap-

plied to CBIR (Smith and Chang, 1994; Do and Vet-

terli, 2002). In general, any measures that provide

some degree of class separation should be included in

the feature set. However, as more features are added,

there is a trade-off between classiﬁcation performance

and computational complexity. In this work, three dif-

ferent feature sets derived from the compressed ECG,

denoted by FS1, FS2, and FS3, are presented and in-

vestigated.

We began by using the subband energies as a ﬁrst

step towards an efﬁcient characterization of texture

in ECG images. It has been suggested (Smith and

Chang, 1994) that the texture content of images can

be represented by the distribution of energy along

the frequency axis over scale and orientation. For

each subband, the dequantized wavelet coefﬁcients

ˆs

(m,n) are computed as

ˆs

(m,n) = {v

(m,n) + δ · sign[v

(m,n)]} · ∆

, (2)

where δ ∈ [0,1) is a user deﬁned bias parameter. Al-

though the value of δ is not normatively speciﬁed in

the standard, it is likely that many decoders will use

the value of one half. Then, the energy of subband j

is deﬁned as

∑

m=1

∑

n=1

ˆs

(m,n). (3)

The resulting subband energy-based features are used

as a morphological descriptor of ECG signals. An-

other feature of interest is the average time elapse

between two successive R peaks, referred to as the

. Certain ectopic heartbeats, such as premature

ventricular contraction and atrial premature beats, are

related with premature heartbeats that provide shorter

RR-intervals than other types of ECG signals. Thus,

changes in the RR-interval plays an important role

in characterizing the dynamics information around

the heartbeats. Notice that the RR

can be calcu-

lated from the heartbeat lengths which are transmitted

as side information along with the JPEG2000 code-

stream. With J-level wavelet decomposition, a to-

tal of (3J + 2) features are used to form a biomet-

ric identiﬁcation vector (BIV) of the subject. The

BIV used for the FS1 will be denoted as b

(1)

,...,E

3J+1

,RR

The second feature set FS2 is obtained by ap-

plying principal component analysis (PCA) (Bishop,

2006) on wavelet coefﬁcients from the lowpass sub-

band S

. The validity of using S

is supported by

the fact that the lowpass subband represents the ba-

sic ﬁgure of an image, which features a high sim-

ilarity among the ECGs of the same person. In

order to achieve dimension reduction, PCA ﬁnds

projection vectors in the directions of highest vari-

ability such that the projected samples retain the

most information about the original data samples.

To begin, consider a training dataset consisting of

ECG images of dimension 200 × 200. Let M

and

denote the row and column dimensions of the

lowpass subband S

after the wavelet decomposi-

tion, respectively. Before applying the PCA, M

rows of the subband S

are concatenated to form a

wavelet coefﬁcient vector of length M

× N

, i.e.,

u = {ˆs

(1,1), ˆs

(1,2),..., ˆs

)}. Then, the

mean vector

u and the covariance matrix Σ

are com-

puted using the training dataset. Following the eigen-

decomposition, we obtain an eigenvalue matrix D and

its corresponding eigenvector matrix V with its col-

umn vectors sorted in the descending order of eigen-

values. Finally, the wavelet vector u is projected

on the eigenvectors to obtain the principal compo-

nent vector p = V

(u −

u). In this work, only the

ﬁrst seven principal components are selected based

on containment of 99% of the total variability. To-

gether with the RR-interval, the BIV for the FS2 is

SIGMAP2015-InternationalConferenceonSignalProcessingandMultimediaApplications

composed of eight features and is denoted as b

(2)

{p(1),..., p(7),RR

}. As for the feature set FS3, it

is a combined version of FS1 and FS2 and the corre-

sponding BIV is denoted as b

(3)

= {b

(1)

(2)

3.2 Enrollment and Recognition

Enrollment and recognition are two important stages

of the person identiﬁcation system. In the enrollment

stage, BIVs of each subject are taken as representa-

tions of the subject and enrolled into a database. In

the recognition stage, the query BIV of an unknown

subject is compared with the enrollment database to

ﬁnd a match. The recognition performance depends

on the underlying classiﬁer used for person identi-

ﬁcation. First, nearest-neighbor (NN) classiﬁers are

exploited for testing various feature sets in discrimi-

nating different subjects. The NN classiﬁer tends to

search for the most similar class to a given query BIV

with similarity deﬁned by the normalized Euclidean

distance (Smith and Chang, 1994).

Another classiﬁcation approach considered here

is the support vector machine (SVM) which has

shown effective in many pattern recognition problems

(Bishop, 2006). This is partly due to their good gen-

eralization capability derived from the structural risk

minimization principle. Since person identiﬁcation

involves the simultaneous discrimination of several

subjects, we considered the one-against-one method

for solving multiclass SVM problems (Bishop, 2006).

For a K-class problem, the method constructs K(K −

1)/2 binary SVM classiﬁers where each one is trained

on the training data from two classes. Training the

binary SVM consists of ﬁnding a separating hyper-

plane with maximum margin and can be posed as the

quadratic optimization problem. For the t-th ECG

image, suppose that the pair (x

) contains the fea-

ture vector x

∈ {b

(1)

(2)

(3)

} and its corresponding

class label y

∈ {1,2,...,K}. Given a set of T training

data pairs {(x

),t = 1, 2, . . . , T } from classes i and

j, SVM algorithm can be formulated as the following

primal quadratic optimization problem

min

i j

,ξ

i j

||w

i j

∑

t=1

i j

subject to: (w

i j

)

+ b

i j

≥ 1 − ξ

i j

, if y

= i,

i j

)

+ b

i j

≤ −1 + ξ

i j

, if y

= j,

i j

≥ 0, (4)

where C is a regularization parameter, w

i j

, b

i j

and ξ

i j

are the weight vector, bias and slack variable, respec-

tively. Due to various complexities, a direct solution

to minimize the objective function (4) with respect to

i j

is usually avoided. A better approach is to apply

the method of Lagrange multipliers to solve the dual

problem as follows:

max

i j

∑

t=1

i j

−

∑

t=1

∑

i j

subject to:

∑

t=1

i j

= 0,

0 ≤ a

i j

≤ C, (5)

where a

i j

= (a

i j

,...,a

i j

) is the vector of Lagrange

multipliers. In this work, the dual problem to SVM

learning is solved using the sequential minimal opti-

mization method (Bishop, 2006). After obtaining the

optimum values of Lagrange multipliers a

i j

, it is then

possible to determine the corresponding weight vec-

tor w

i j

and the decision function f

i j

) as follows:

i j

∑

t=1

i j

, (6)

i j

) = (w

i j

)

+ b

i j

. (7)

Finally, the class label y for a query BIV x of a new

subject is determined based on the max-wins voting

strategy (Bishop, 2006). More precisely, each binary

SVM casts one vote for its preferred class and the ﬁnal

result is the class with the highest vote, i.e.,

y = arg max

∑

j6=i, j=1

sign[ f

i j

(x)]. (8)

4 EXPERIMENTAL RESULTS

Computer simulations were conducted to evaluate the

performances of the proposed ECG biometric sys-

tem for both healthy and diseased subjects. ECG

records from the PhysioNet QT Database (Laguna

et al., 1997) were chosen to represent a wide vari-

ety of QRS and ST-T morphologies. First, 10 healthy

subjects that are originally from the MIT-BIH Normal

Sinus Rhythm Database are used in the experiments

and denoted as dataset D1. Subjects that are added to

the database to examine the effects of diseased ECG

consist of 10 records from the MIT-BIH Arrhythmia

Database, 10 records from the MIT-BIH Supraven-

tricular Arrhythmia Database, and 10 records from

the Sudden Cardiac Death Holter Database. These

three groups of diseased subjects are denoted as D2,

D3, and D4, respectively. Each of these records

is 15 minutes in duration and are sampled at 250

Hz with a resolution of 11 or 12 bits/sample. The

JPEG2000 simulation was run on the open-source

CompressedDomainECGBiometricIdentificationusingJPEG2000

Table 1: Average CR and PRD (%) performances for JPEG2000 with coding rate ρ = 0.15 and ρ = 0.08.

Rate ρ = 0.15 Rate ρ = 0.08

Dataset N

50 100 150 200 50 100 150 200

CR 12.19 13.45 14.04 14.28 21.62 21.90 21.84 21.84

PRD 3.59 3.19 3.19 3.08 8.72 5.91 5.55 5.55

CR 9.03 9.37 9.49 9.70 16.81 16.92 16.84 16.84

PRD 4.19 3.48 3.36 3.26 11.36 7.95 7.18 7.18

CR 10.07 10.06 10.06 10.61 18.90 19.19 18.87 18.87

PRD 4.53 3.56 3.45 3.18 11.86 8.75 7.99 7.94

CR 10.11 10.23 10.42 10.05 19.11 19.13 19.13 19.13

PRD 4.21 3.32 3.19 3.29 12.35 8.76 7.92 7.92

software JasPer version 1.900.0 (Adams, 2002). Each

ECG image is regarded as a single tile and the dimen-

sion of the code-block is 64 × 64. ECG images were

compressed in lossy mode using Daubechies 9/7 ﬁl-

ter with 5-level wavelet decomposition. Besides, the

targeted coding rate ρ was empirically determined to

be 0.15 and 0.08 in order to achieve the compression

ratio in the region of 10 and 20, respectively.

A preliminary experiment was ﬁrst conducted to

examine the performance dependence of 2-D com-

pression on the number N

of heartbeats employed in

constructing an ECG image. The system performance

is evaluated in terms of the compression ratio (CR)

and the percent root mean square difference (PRD).

The CR is deﬁned as CR = N

ori

com

, where N

ori

and

com

represent the total number of bits used to code

the original and compressed ECG data, respectively.

The PRD is used to evaluate the reconstruction distor-

tion and is deﬁned by

PRD(%) =

∑

l=1

ori

(l) − x

rec

(l)]

∑

l=1

ori

(l)

× 100, (9)

where L is the total number of original samples in

the record and x

ori

and x

rec

represent the original

and reconstructed ECG signals, respectively. Table

1 presents the average results of CR and PRD for

JPEG2000 simulation with coding rate ρ = 0.15 and

ρ = 0.08. As expected, the system yielded better

performance with an increase in the heartbeat num-

ber N

. However, the increasing delay may limit its

practical applicability as the JPEG2000 must buffer a

total of N

× 200 samples before it can start encod-

ing. Thus, we empirically choose N

= 200 as the

best compromise in the sequel. Note that while the 2-

D compression method works well for normal ECGs,

it may suffer from ECG irregularities due to the false

QRS detection in the preprocessor stage. For the case

of ρ = 0.08 and N

= 200, the PRD performance has

dropped from 5.55% in the D1 to 7.92% in the D4.

In order to justify the efﬁciency of the proposed

method, we also analyze the run-time complexity

of JPEG2000 decoder for ECG data. According to

the JPEG2000 coding standard, its full decompres-

sion process can be highlighted as: entropy decod-

ing, dequantization to obtain the DWT coefﬁcients,

and inverse DWT to reconstruct blocks of pixels. By

studying the code execution proﬁles, we can see that

the decoder spends most of its time on the inverse

DWT (typically 61.5% or more). By contrast, the

amount of time consumed by entropy decoding and

de-quantization is about 30.8%. This observation is

in accord with the results for natural and synthetic im-

agery produced by the Jasper software implementa-

tion, reported earlier (Adams and Kossentini, 2000).

It was found that the execution time breakdown for

the JPEG2000 decoder depends heavily on the par-

ticular image and coding scenarios employed. In

the case of lossless image compression, tier-1 decod-

ing usually requires the most time (typically more

than 50%), followed by the inverse DWT (typically

30 − 35%), and then tier-2 decoding. For lossy im-

age compression, as in our case, the inverse DWT ac-

counts for the vast majority of the decoder’s execution

time (typically 60 − 80%). The complexity analysis

results demonstrate that the proposed method has the

advantage of by-passing the inverse DWT operation.

The next step is to evaluate the recognition per-

formances of NN classiﬁers on the feature sets FS1

and FS2. Both feature sets have in common the RR

to provide dynamics feature of ECG signals. Mor-

phological features to be computed for the FS1 are

subband energies {E

,1 ≤ j ≤ 16}, and PCA-based

features {p(i),1 ≤ i ≤ 7} for the FS2. The system per-

formance is evaluated in terms of the recognition rate,

which is normally deﬁned as the ratio of the num-

ber of correctly identiﬁed subjects to the total num-

ber of testing subjects. First of all, the proposed sys-

tems were individually tested on datasets from D1

to D4. Since ECG records from the QT database

may vary in the number of heartbeats, a total of 4

to 8 ECG images of dimension 200 × 200 would be

generated for different individuals. For a fair assess-

ment, 1000 trials of repeated random sub-sampling

were implemented to eliminate possible classiﬁcation

SIGMAP2015-InternationalConferenceonSignalProcessingandMultimediaApplications

biases. In each trial, four compressed ECG images

per subject were randomly selected for feature extrac-

tion. Among them, the ﬁrst two ECG images are used

for training in the enrollment stage, and the other two

are used for testing in the recognition stage. Table 2

presents the average recognition rates associated with

various datasets for JPEG2000 coding with ρ = 0.15

and ρ = 0.08. Compared with the subband energy-

based FS1, the improved performances of FS2 indi-

cate that morphological features of ECG signals are

better to be exploited in the DWT domain. The re-

sults also show that the recognition performances are

affected by ECG variations caused by cardiovascu-

lar diseases. For the case of FS2 and ρ = 0.15, the

recognition rate has dropped from 100% for D1 to

91.77% for D4. The possible reasons resulting in a

lower recognition rate for diseased subjects could be

unstable QRS-complex that appear aperiodically in

ECGs. It is important to note that the lower value

of ρ = 0.08 did not result in a signiﬁcant performance

degradation. This can be attributed to the fact that

the JPEG2000 supports a lower coding rate by en-

larging the quantizer step sizes within high frequency

subbands. This has little effect on the recognition

performance, as the frequency content of the QRS

complex is most concentrated in low frequency sub-

bands. To elaborate further, we also include in Ta-

ble 2 the recognition performances for the case where

normal and diseased subjects are jointly enrolled and

tested. As the table shows, the additional inclusion

of 30 diseased subjects has dropped the recognition

rate by 4.45%. The results indicate that the com-

bined use of FS2 and NN classiﬁer is still not robust

enough to handle the inclusion of diseased patients

in the database. In addition to the above-mentioned

schemes, we also examine how the recognition per-

formance changes as a function of the number of

heartbeats used in constructing an ECG image. The

results are illustrated in Figure 4 for various image

sizes using an NN classiﬁer and a combined dataset

from D1 to D4. Our general conclusion is that better

performances can be achieved with an increase in the

heartbeat number N

, but the performance has a ten-

dency to become ﬂattened for N

≥ 100. This would

be beneﬁcial for future works on real-time implemen-

tation of the proposed ECG biometric systems.

Another problem which requires further investi-

gation is to test the proposed system for the situa-

tion where subjects were identiﬁed solely by means

of DWT-based morphological features. The perfor-

mances of NN classiﬁers for FS1 and FS2 without in-

corporating the RR-interval are summarized in Table

3. A comparison between Table 2 and 3 indicates that

the inclusion of RR-interval is likely to help in dis-

Table 2: Recognition rates (%) of the feature sets FS1 and

FS2 using an NN classiﬁer.

Dataset

ρ = 0.15 ρ = 0.08

FS1 FS2 FS1 FS2

96.16 100 95.29 99.98

86.82 91.82 86.73 91.75

91.61 95.62 91.75 95.50

84.50 91.77 83.52 91.91

D1, D2, D3, D4

88.92 95.55 88.85 95.57

Figure 4: Recognition rates for different ECG image sizes

using JPEG2000 with ρ = 0.15.

criminating between subjects with greater accuracy.

However, we should state that the issue of whether

it is better to use RR-interval in the ECG biomet-

ric does not appear to be settled and may be prob-

lem dependent (Odinaka et al., 2012). Prior works on

ECG biometric (Biel et al., 2001; Israel et al., 2005),

(Suﬁ and Khalil, 2011) have suggested the use of the

RR-interval as a second source of biometric data to

supplement morphological features of ECG signals.

However, almost all these works exploited a database

containing ECG signals in rest conditions, which rep-

resents also a limitation for practical issues of bio-

metric systems. Recent studies (Odinaka et al., 2012;

Poree et al., 2014) have shown that ECG signals ac-

quired in different physiological conditions allow for

substantial variability in heart rates, which could sig-

niﬁcantly impact the performance of the biometric

system. Continuing this research, we will address

ourselves to the study of robust ECG biometric al-

gorithms utilizing biometric features without the RR-

interval.

We next compare the performances of NN and

SVM classiﬁers trained on the combined feature set

FS3. With 5-level wavelet decomposition, a total of

24 features are used to form the FS3, including RR

{p(i),1 ≤ i ≤ 7}, and {E

,1 ≤ j ≤ 16}. Table 4 gives

the recognition performances of various datasets us-

ing NN and SVM for classiﬁcation. With respect to

the implementation of SVM classiﬁers, the simula-

CompressedDomainECGBiometricIdentificationusingJPEG2000

Table 3: Recognition rates (%) of NN classiﬁers using feature sets FS1 and FS2 without RR-interval.

Dataset

JPEG2000 (ρ = 0.15) JPEG2000 (ρ = 0.08)

FS1 FS2 FS1 FS2

(no RR

) (no RR

)

95.77 99.73 94.93 99.65

85.61 88.18 85.93 88.40

91.27 95.34 90.84 95.57

80.98 90.20 77.73 89.69

D1, D2, D3, D4

86.32 94.58 86.46 94.56

Table 4: Recognition rates (%) of the feature set FS3 using

NN and SVM classiﬁers.

Dataset

ρ = 0.15 ρ = 0.08

NN SVM NN SVM

100 100 99.99 100

98.99 100 98.93 100

99.85 100 99.99 100

97.37 100 97.44 100

D1, D2, D3, D4

99.09 100 99.15 100

QT database

96.06 99.50 96.15 99.48

tion was run on the open-source software LIBSVM

(Chang and Lin, 2011), which supports SVM formu-

lations for various classiﬁcation problems. The re-

sults clearly demonstrate the improved performances

achievable using the FS3 in comparison to those of

FS1 and FS2, even with a simple NN classiﬁer. The

main reason for this is that the FS3 features a high

similarity among the BIVs of the same person, ei-

ther healthy or with CVD, but a much lower simi-

larity between two BIVs of different persons. The

results also indicate that the SVM classiﬁers trained

on FS3 are very effective with 100% recognition rate

in all test datasets. Compared with NN classiﬁer,

the SVM shows better generalization ability on the

limited training data, which is indeed the case per-

formed in this study. Also included in the table are

the performances when 86 ECG records from the QT

database are jointly enrolled and tested. The proposed

system can achieve 99.50% and 99.48% in accura-

cies for JPEG2000 compression with ρ = 0.15 and

ρ = 0.08, respectively. The results indicate that the

combined use of FS3 and SVM classiﬁer is more in-

variant to ECG irregularities induced by cardiovascu-

lar diseases.

5 CONCLUSIONS

This paper proposed a robust method for biometric

identiﬁcation using wavelet-based features extracted

from the JPEG2000 compressed ECG. Under the

JPEG2000 framework, the person identiﬁcation prob-

lem is analogous to a content-based image retrieval

problem. Morphological features of ECG signals

are derived directly from the DWT coefﬁcients with-

out involving full decompression of JPEG2000 bit-

stream. Also, dynamic feature such as RR-interval

proved to be beneﬁcial for ECG biometric. Combined

performance of both healthy persons and diseased pa-

tients indicate that the proposed system enables faster

biometric identiﬁcation in compressed domain and in

the same time it is more invariant to ECG irregulari-

ties induced by cardiovascular diseases.

It should be noted that the results presented in this

work, while promising, were obtained from a mod-

erate size of datasets. Also, the experiments were

conducted off-line, using ECG signals collected under

controlled conditions from public databases. In prac-

tical applications, however, a range of issues would

require further investigations. First, the ECG biomet-

ric system needs to be tested in more realistic environ-

ments, varying with respect to the type and quantity of

data collected. Second, the effects of varying mental

and physiological conditions on the recognition accu-

racy, as well as the delay induced by JPEG2000 cod-

ing also need to be resolved.

ACKNOWLEDGEMENTS

This research was supported by the National Science

Council, Taiwan, ROC, under Grant NSC 102-2221-

E-009-030-MY3.

REFERENCES

Aboy, M., Crespo, C., McNames, J., Bassale, J., Jenkins,

L., and Goldsteins, B. (2002). A biomedical signal

processing toolbox. In Biosignal’02. (pp. 49-52).

Adams, M. D. (2002). The JasPer Project Home

Page. Retrieved may 22, 2015, from http://www.ece.

uvic.ca/∼frodo/jasper/.

Adams, M. D. and Kossentini, F. (2000). JasPer: A

software-based JPEG-2000 codec implementation.

In International Conference on Image Processing.

(pp. 53-56). IEEE.

SIGMAP2015-InternationalConferenceonSignalProcessingandMultimediaApplications

Agraﬁoti, F. and Hatzinakos, D. (2009). ECG biometric

analysis in cardiac irregularity conditions. Signal, Im-

age, Video Process., 3(4):329–343.

Biel, L., Pettersson, O., Philipson, L., and Wide, P. (2001).

ECG analysis: A new approach in human identiﬁca-

tion. IEEE Trans. Instrum. Meas., 50(3):808–812.

Bilgin, A., Marcellin, M. W., and Altbach, M. I.

(2003). Compression of electrocardiogram signals

using JPEG2000. IEEE Trans. Consum. Electron.,

49(4):833–840.

Bishop, C. M. (2006). Pattern Recognition and Machine

Learning. Springer, New York.

Chang, C. C. and Lin, C. J. (2011). LIBSVM: A li-

brary for support vector machines. ACM Trans. In-

tell. Syst. Techn., 2(3):27.

Chiu, C. C., Chuang, C. M., and Hsu, C. Y. (2009). Dis-

crete wavelet transform applied on personal identity

veriﬁcation with ECG signal. Int. J. Wavelets Multi.,

7(3):341–355.

Daubechies, I. and Sweldens, W. (1998). Factoring wavelet

transforms into lifting steps. J. Fourier Anal. Appl.,

4(3):245–267.

Do, M. N. and Vetterli, M. (2002). Wavelet-based tex-

ture retrieval using generalized gaussian density and

kullback-leibler distance. IEEE Trans. Image Pro-

cess., 11(2):146–158.

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

and Wiederhold, B. K. (2005). ECG to identify indi-

viduals. Pattern Recognit., 38(1):133–142.

Jalaleddine, S. M., Hutchens, C. G., Strattan, R. D.,

and Coberly, W. (1990). ECG data compres-

sion techniques - a uniﬁed approach. IEEE

Trans. Biomed. Eng., 37(4):329–343.

Laguna, P., Mark, R. G., Goldberger, A. L., and Moody,

G. B. (1997). A database for evaluation of algorithms

for measurement of QT and other waveform inter-

vals in the ECG. In Computers in Cardiology 1997.

(pp. 673-676). IEEE.

Linear Dimensions Inc. (2015). Biometric ECG secu-

rity devices. Retrieved may 22, 2015, from http://

www.lineardimensions.com/securitycardﬁnal.pdf.

Nymi Inc. (2013). Nymi white paper. Retrieved may

22, 2015, from https://www.nymi.com/wp-content/

uploads/2013/11/nymiwhitepaper-1.pdf.

Odinaka, I., Lai, P. H., Kaplan, A. D., OSullivan, J. A.,

Sirevaag, E. J., and Rohrbaugh, J. W. (2012). ECG

biometric recognition: A comparative analysis. IEEE

Trans. Inform. Forensic Secur., 7(6):1812–1824.

Plataniotis, K. N., Hatzinakos, D., and Lee, J. K. (2006).

ECG biometric recognition without ﬁducial detection.

In BSYM’06, Biometrics Symposium: Special Session

on Research at the Biometric Consortium Conference.

(pp. 1-6). IEEE.

Poree, F., Kervio, G., and Carrault, G. (2014). ECG biomet-

ric analysis in different physiological recording condi-

tions. J. Signal, Image and Video Process.

Smith, J. R. and Chang, S. F. (1994). Transform features for

texture classiﬁcation and discrimination in large im-

age databases. In Int. Conf. Image Process. (pp. 407-

411). IEEE.

Suﬁ, F. and Khalil, I. (2011). Faster person identiﬁca-

tion using compressed ECG in time critical wireless

telecardiology applications. J. Netw. Comput. Appl.,

34(1):282–293.

Sun, C. C. and Tai, S. C. (2005). Beat-based ECG com-

pression using gain-shape vector quantization. IEEE

Trans. Biomed. Eng., 52(11):1882–1888.

Taubman, D. S. and Marcellin, M. W. (2001). JPEG

2000: Image Compression Fundamentals, Standards,

and Practice. Kluwer Academic Publishers, Norwell,

MA.

Thaler, M. S. (2012). The Only EKG Book You’ll Ever Need.

Lippincott Williams & Wilkins, Philadelphia, 7th edi-

tion.

Yang, M., Liu, B., Zhao, M., Li, F., Wang, G., and Zhou,

F. (2013). Normalizing electrocardiograms of both

healthy persons and cardiovascular disease patients

for biometric authentication. PloS one, 8(8):e71523.

CompressedDomainECGBiometricIdentificationusingJPEG2000