ECG-based Human Authentication: A Review

Dhiraj Ramnani1, Shail Patil2, Darpen Patel3, Prof. Hetal Gaudani4

Student1,2,3, Associate Professor4
1,2,3,4
Computer Engineering Department, G H Patel College Of Engineering And Technology
Gujarat Technological University, Gujarat, India

Abstract- The electrocardiogram (ECG) is an emerging methods use latency, slope, angle and other measurements
novel biometric approach for human identification. This derived from anchor points within the signal (e.g. P-QRS-T
project investigates the possibility of biometric human complexes), to create feature vectors that are used as input to
identification based on the electrocardiogram (ECG). The the recognizer [19, 26]. Partially-fiducial methods typically
ECG, being a record of electrical currents generated by the only use the R-peak to perform heartbeat waveform
heart, is potentially a distinct human characteristic, since segmentation, adopting either the full waveform or a subset of
ECG waveforms and other properties of the it as input to the recognizer [23]. Non- fiducial methods extract
electrocardiogram depend on the anatomic features of the information from the structure of the signals, without the need
human heart and body. Many aspects of our everyday lives for any reference points [23,27] these methods are especially
are becoming dependent on automatic and accurate advantageous given that the process of identifying fiducial
identity validation. The vast deployment of recognition points can be challenging.
mechanisms, based on something that people have (entity- 1. Fiducial methods- It depends on local features of ECG
based: tokens and ID cards) or something that people know signal for biometric template design, such as temporal or
(knowledge-based: PIN numbers and passwords), raises amplitude difference between consecutive fiducial points and
security concerns. The major benefit of security systems used for identification purposes. The drawback of fiducial
based on biometrics is the full dependency on the features is their sensitivity to noise. Furthermore, detection of
individual. The main purpose of this study is to present a fiducial features in unusual cases with arrhythmia may include
novel personal authentication approach for human errors in data.[31]
authentication. 2. Non-fiducial method- It was initially proposed by
Plataniotis et al. [28,31] in order to remove the necessity of
Keywords- Human Authentication, Biometrics, Feature fiducial point localization of the ECG signal. It treats ECG
Extraction, Security signal or secluded heartbeats holistically and captures features
based on the overall morphology of waveform [28,31].
I. INTRODUCTION
II. ECG BASICS
The potential of ECG as a powerful authentication entity has
been lately explored in great depth owing to its characteristics The human electrocardiogram reflects the specific pattern of
of human anatomy which are almost implausible to replicate electrical activity of the heart throughout the cardiac cycle and
and its benefit of persistence and aliveness of the entity. Its can be seen as changes in potential difference. The ECG is
prospective use as a biometric, surpassing its sole use of affected by a number of physiological factors including age,
clinical diagnosis can be justified by the aspects like: body weight, and cardiac abnormalities.
 Uniqueness-there are no two individual with identical Electrocardiogram (ECG) recordings are quantitative
characteristics, measures of the electrical activity of a person's heart over
 Universal-it can be virtually found in every human being, time[18]. It consists of P-wave, QRS-wave, and T-wave. The
 Measurability-it can be easily acquired using suitable QRS wave is usually called QRS composite or QRS complex,
devices, and it contains the most significant information in that
 Permanence-does not vanish over time, heartbeat. A typical beat in an electrocardiogram comprises of-
 Performance-it has been shown to provide accurate results
for the subset population. [23, 30].
1. A low amplitude P-wave, representing atrial depolarization.
The study of ECG and obtaining relevant and consistent data
2. The QRS complex of much higher amplitude than the P-
from it and matching it with pre-calculated numerics stored in
wave, representing ventricular depolarization.
the database is the key to its success. Extraction of unique 3. A T-wave of smaller amplitude and larger duration than the
features of ECG is a cumbersome task which gives the QRS complex, representing ventricular repolarization.[25]
biometric cutting edge accuracy and uniqueness.
Regarding Template Extraction, existing approaches can be
defined as fiducial, partially fiducial and non-fiducial. Fiducial
 b. Preprocessing of ECG
A. ECG
The raw ECG is rather noisy and contains distortions which
may reduce the effectiveness of feature extraction and
classification and thus lead to the decrease in efficiency of the
results produced and also the accuracy.
● Digital filtering: It is the primary step of preprocessing. It is
used to filter unnecessary noise in ECG signals. Example the
digital notch filter is employed to remove the power line
interference. A high pass band filter with a cut-off frequency of
0.5 Hz is used to remove the DC offset on the obtained
signal.
 Down Sampling: It is a consistent solution for algorithms
performance on high sampled signals.
 Peak detection: In peak detection, the unprocessed signal
is digitally filtered and the peak index detected is used in
the feature selection setup.
Figure 1: ECG complex[25,29]  Segmentation: In this step, a particular peak of the signal
is identified first then ECG waveform is segmented into an
ECG is a method to measure and record different electrical individual heartbeat.[29]
potentials of the heart. The ECG may roughly be divided into
the phases of depolarization and repolarization of the muscle c. Feature Extraction
fibers making up the heart. The depolarization phases
correspond to the P-wave (atrial depolarization) and QRS-wave After pre-processing, the second stage towards classification is
(ventricular depolarization). The repolarization phases to detect certain features of ECG signals mostly QRS complex,
correspond to the T-wave and U-wave (ventricular P and T waves. Statistical features like Dower matrix,
repolarization). The basic elements in the ECG-complex are correlation coefficient, covariance matrix etc, wavelet features
shown in Fig 1[19, 25]. like ICA, Fourier transform, Discrete Cosine Transform (DCT)
etc. techniques are used to extract the features of ECG[25,29].
B. BASIC STEPS INVOLVED IN ECG However, the initial features may include redundant and
useless information due to correlation and interdependencies
The steps involved are as follows: between features. Two methods of feature space reduction are
1. Signal Acquisition proposed: Principal Component Analysis (PCA) and Wavelet
2. Preprocessing Transform (WT) [30].
3. Feature extraction
4. Classification [25] d. Classification

The goal of classification is to identify a subject or to verify an

identity claim from the sensor observations [29]. Various
classification algorithms namely Radial Basis Function (RBF),
K Nearest Neighbor (kNN), Bayes Network (BN), Multilayer
Perceptron (MLP), soft independent modeling of class analogy
(SIMCA), Principle component analysis (PCA), are described
in literary works [25, 29].

Figure 2: Process Flow for ECG processing [29] III. COMPARISON

a. Signal Acquisition The given table performs a comparison between different

approaches used in the referred research papers.
The initial step for entire authentication is acquiring the It compares the following factors:
signals using sensors or other hardware devices. The data 1. Features
collected is either processed or raw data is stored in the 2. Feature Extraction Methods
database which is later used to match the testing dataset for 3. No. of subjects
authenticating a valid user. 4. Accuracy
 Paper Studied Features Feature Selection No. of subjects Accuracy
Methods used

Sebastian et al. Fiducial Template Matching 73 records 81.82%

[14] (Physionet
database)
Choudhary et al. Amplitude of QRS Discrete Cosine 3969 imposters FRR-11.6%
[15] complex Transform (DCT) and and 19404 test FAR-5.8%
R-peak alignment Template Matching segments Detection
Accuracy-
100%
Safie et al. The peak of P to Multiple Pulse k- 20 Classificati
[16] the peak of T (P2T) Nearest Neighbor on-100%
(MPkNN)
Kumar et al. ICA components Wavelet Transform The proposed Recognition
[18] and wavelet (WT) and Independent method was 99.6%
coefficients Component Analysis tested on three
(ICA) public ECG
databases.[1]
Biel et al. Temporal, Correlation matrix 20 100%
[19] amplitude and
slope
Belgacem et al. Amplitude and Discrete Wavelet 20 100%
[21] normalized time Transform &
distances between Random Forest
successive fiducial Algorithm-
points. Authentication
Singh et al. Fiducial points Template Matching 50 98%
[22] P-QRS-T
Complexes
Fred et al. Partial Fiducial Template Matching & 64 EER- 9.1%
[23] Approach(Use of R SVM
wave to perform
segmentation)
Shen [13] Fiducial Template Matching + 20 100%
DBNN

Israel [11] Fiducial LDA 29 98%

Chiu[12] Non Wavelet 35 100%

Fiducial Distance+LDA

Janani, David[9] Both KNN + Bayesian 17 88%

Table 1. Comparison of methodology [25]

 IV. EXISTING LITERATURE

The features, in most biometric methods, were extracted in the correct verification rate was 95% for template matching, 80%
direct time-domain, frequency-domain and time-frequency- for the DBNN, and 100% for combining the two methods[25].
domain representation of the ECG signal. The spectral features Shen [13] extended the proposed methods in a larger database
are extracted from the Fourier transform (FT) of time-domain that contained 168 healthy subjects. Template matching and
ECG signal or decomposed ECG signals. The performance of mean square error (MSE) methods were compared for
different methods depends on classifiers such as K-nearest prescreening, and distance classification and DBNN compared
neighbor (KNN), back-propagation neural network (BPNN), for second-level classification. The features used for the
hidden Markov model (HMM), Bayes Network (BN), Naive second-level classification are seventeen temporal and
Bayes (NB), and support vector machine (SVM). Certain amplitude features. The best identification rate for 168 subjects
methods are based on feature analysis, using one or more is 95.3% using template matching and distance
techniques such as LDA, principal component analysis (PCA), classification.[25]
independent component analysis (ICA), random projection Wang et al. [8] made the initial efforts that did not depend on
(RP) in conjunction with K- SVD based dictionary learning, fiducial based features and relied on the combination of a set of
and within-class covariance normalization (WCCN) for analytic features derived from Fiducial points with appearance
dimensionality reduction. The test was conducted using the features obtained using PCA and LDA(principal component
ECG signals taken from the well-known ECG databases: MIT- analysis and linear discriminate analysis) for feature extraction
BIH arrhythmia (MITADB), MIT-BIH supra-ventricular and data reduction. The accuracy for 13 subjects was 84%
arrhythmia database (MITSADB), MIT-BIH Normal sinus using analytic features alone and 96% using LDA with K-
rhythm (NSRDB), Physikalisch-Technische Bundesanstalt NN(K-nearest neighbor). The combination of the types of
(PTB) and European ST-T (ESTTDB) [1]. features was used to achieve 100% accuracy. Janani, et al [9]
In this section, we provide a review of the related literary handled activity induced ECG variation by extracting a set of
works on ECG-based human authentication. Biel et al. [3] are accelerometer features that characterize different physical
among the earliest initiatives that demonstrate the plausibility activities along with fiducial and non fiducial ECG features.
of utilizing ECG for human identification purposes. A set of Chan et al. [10] proposed another non-Fiducial feature
temporal and amplitude features are extracted from a extraction framework using a set of distance measures
SIEMENS ECG equipment directly. A feature selection including a novel wavelet transform distance. Data was
algorithm based on simple analysis of correlation matrix is collected from 50 subjects using button electrodes held
used to reduce the dimensionality of features. A multivariate between the thumb and finger. The wavelet transform distance
analysis based method is employed for classification. The outperforms other measures with an accuracy of 89%.[25]
system was tested on a database of 20 individuals, and 100% Sebastian et al. [14] proposed biometric authentication
identification rate was achieved, based on empirically selected algorithm for mobile devices. In this approach, the user touches
features. A major limitation of Biel et al.'s method is the lack two ECG electrodes (lead I) of the mobile device to gain
of automatic recognition due to the usage of specific equipment access. The algorithm was tested with a cell phone case heart
for feature extraction. This declines the scope of applications. monitor in a controlled laboratory experiment at different times
Irvine et al. [26] proposed a system to use heart rate variability and conditions with ten subjects and also with 73 records
(HRV) as a biometric for human identification. Israel et al. [11] obtained from the Physionet database. The obtained results
proposed a more extensive set of descriptors to characterize reveal that our algorithm has 1.41% false acceptance rate and
ECG signal. An input ECG signal is first preprocessed by a 81.82% actual acceptance rate.
bandpass filter. The peaks are established by finding the local Choudhary et al. [15] presented a simple unified accumulation
maximum in a region surrounding each of the P, R, T of averaged heartbeat extraction framework for noise-robust
complexes, and minimum radius curvature is used to find the ECG-based biometric authentication using pulse morphology.
onset and end of P and T waves. A total number of 15 features, There are three major steps involved in the proposed method:
which are time duration between detected fiducial points, are preprocessing, combined averaged beat construction, and
extracted from each heartbeat. This system was tested on a similarity matching. The signal blocking and mean removal
database of 29 subjects with 100% human identification rate operations are performed at the preprocessing stage. The
and around 81% heartbeat recognition rate can be achieved. ensemble averaging step includes discrete cosine transform
Shen et al. [13] introduced a two-step scheme for identification (DCT) based filter for simultaneous removal of BW and PLI
from one-lead ECG. A template matching method is first noises, straightforward Gaussian derivative filter (GDF)-based
employed to compute the correlation coefficient for R-peak detector, peak-centering and period normalization, and
comparison of two QRS complexes. A decision-based neural ensemble averaging computation.A modified version of the k-
network (DBNN) approach is then applied to complete the Nearest Neighbor (kNN) classifier named Multiple Pulse k-
verification from the plausible candidates selected with Nearest Neighbor (MPkNN) was introduced by Safie et al.
template matching. The inputs to the DBNN are seven [16].Combining the distinctive characteristic of the Pulse
temporal and amplitude features extracted from QRS-T wave. Active Width (PAW) technique with kNN led to the
The experimental results from 20 subjects showed that the development of MPkNN. This approach was implemented for
class attendance system. Kumar et al. [18] used Wavelet REFERENCES
Transform (WT) and Independent Component Analysis (ICA)
methods to extract morphological features that appear to offer [1] L. Goldberger et al., “PhysioBank, PhysioToolkit, and
sharp distinction among subjects. The proposed method is PhysioNet: Components of a new research resource for
aimed at the two-lead ECG configuration that is routinely used complex physiologic signals,” Circulation, vol. 101, no.
for long-term continuous monitoring of heart activity. To 23, pp. 215-220, 2000.
achieve improved subject identification, the information from
the two ECG leads is fused. [2] Dr. Boo-Ho yang, Prof Haruhiko H. Asada ,Yi Zhang
In [21], the cardiac signals were used to identify a total of 80 “Cuff-less continuous monitoring of beat-to-beat Blood
individuals obtained from four ECG databases from the pressure using a kalman filter and sensor” ,Fusion-first
Physionet database (MIT-BIH, ST-T, NSR, PTB) and an ECG joint IEEE conference of bmes/embs,1999.
database collected from 20 student volunteers from Paris Est. [3] L. Biel, O. Pettersson, L. Philipson, and P. Wide. " ECG
University. Discrete Wavelet Transform (DWT) was used to analysis: a new approach in human identification".
perform Feature extraction. Wavelets have turned out to be Proceedings of the 16th Instrumentation and
particularly effective for extracting distinct features in ECG Measurement Technology Conference, IEEE, volume
signal classification. The Random Forest was then suggested 1,1999.
for the ECG signals authentication. Preliminary experimental [4] Julio C.D Conway,Claudionor J.N Coelho ,Luis C.G
results indicate that the system is accurate and can achieve a Andrade, “Wearable computer as a multi-parametric
low false negative rate, low false positive rate and a 100% monitor for physiological signals” IEEE international
subject recognition rate for healthy subjects with the reduced conference on bioinformatics and biomedical
set of features. In [22], a new authentication strategy is engineering, pp-236-242,2000.
suggested, which uses the outlined features and taking [5] M. Sanjeev Dasrao,Yeo Joon Hock,Eugene KW sim,
identification decisions with respect to the template database “Diagnostic blood pressure wave analysis and
on the basis of match scores. The performance of the system is ambulatory monitoring using a novel”, non-invasive
measured in both a unimodal framework and also in the portable device-proceeding of international conference
multimodal framework where ECG is combined with the face on biomedical engineering, pp 267-272, 2001.
biometric and with the fingerprint biometric. The equal error [6] Peter Varady Benyo ,Balazs, “An open architecture
rate (EER) result of the unimodal system is 10.8%, while the patient monitoring system using Standard technologies”,
EER results of the multimodal systems are reported to 3.02% IEEE transactions on information technology in
and 1.52%, respectively for the systems when ECG is biomedicine, vol. 6, 2002.
combined with the face biometric and ECG integrated with the [7] Paul Lucowicz, Urs Anliker, Jamie Ward, Gerhard
fingerprint biometric. Fred et al. [23] employed the use of Troster, “Amon: a wearable medical computer for high
SVM and template matching for feature selection. The risk patients”, IEEE computer society, 6th international
approach is partially fiducial as it focuses on R-wave symposium on wearable computers (iswc.02), 2002.
segmentation. [8] Y. Wang, F. Agrafioti, D. Hatzinakos, and K. N.
Plataniotis, "Analysis of human electrocardiogram for
biometric recognition", EURASIP Journal on Advances
in Signal Processing, 2008.
V. CONCLUSION [9] Janani S., Minho S., Tanzeem C., David K., ”Activity-
aware ECG-based patient authentication for remote
An exhaustive search of the literature on ECG as a biometric health monitoring”; , International Conference on
for authentication has led to strong proposal of its use because Mobile Systems,ACM,Nov,2009.
its non-mimicable characteristic can provide a more robust and [10] D. C. Chan, M. M. Hamdy, A. Badre, and V. Badee.
efficient human identification system. To increase the ”Wavelet distance measure for person identification
using electrocardiograms”, Transactions on
intricacies of the system, the unimodal ECG can be combined
Instrumentation and Measurement, IEEE, Volume 57,
with other previously explored biometric like finger and retina issue 2, 2008.
scan to develop a multimodal system that is impossible to [11] Steven A. Israel, Jhon M.I,Andrew C.,Mark D.W,”ECG
spoof by a spurious user. The integrity and accuracy of the to identify individuals”, Virtual reality medical center,
system depend on the feature extraction algorithms. The future USA, Pattern Recognization, Volume 38,issue1,2005.
work essentially focuses on fast and correct feature extraction [12] C. Chiu, C. Chuang, and C. Hsu. “A novel personal
identity verification approach using a discrete wavelet
by utilizing more statistical data.
transform of the ECG signal”, Proceedings of the
International Conference on Multimedia and Ubiquitous
Engineering, Volume 6, issue 4,2008.
[13] T. W. Shen, W. J. Tompkins, and Y. H. Hu "One-lead
ECG for identity verification",Proceedings of the 24th
Annual Conference on Engineering in Medicine and
 Biology and the Annual Fall Meeting of the Biomedical using Ziv-Merhav cross parsing. "Pattern Recognition
Engineering Society, Volume 57, issue 2, 2002. (ICPR), 2010 20th International Conference on. IEEE,
[14] Arteaga-Falconi, Juan Sebastian, Hussein Al Osman, 2010.
and Abdulmotaleb El Saddik. "ECG Authentication for [28] Plataniotis, Konstantinos N., Dimitrios Hatzinakos, and
Mobile Devices." IEEE Transactions on Instrumentation Jimmy KM Lee. "ECG biometric recognition without
and Measurement 65.3 (2016): 591-600. fiducial detection." 2006 Biometrics Symposium:
[15] Choudhary, Tilendra, and M. Sabarimalai Manikandan. Special Session on Research at the Biometric
"A novel unified framework for noise-robust ECG- Consortium Conference. IEEE, 2006.
based biometric authentication." Signal Processing [29] Ahuja, Pooja, and Abhishek Shrivastava. "A Review on
and Integrated Networks (SPIN), 2015 2nd International ECG based Human Authentication." (2016)
Conference on. IEEE, 2015. [30] Lugovaya, Tatiana S. "Biometric Human Identification
[16] Safie, Sairul, et al. "Multiple pulse K-Nearest Neighbors Based On ECG." (2005).
authentication for Malay ECG based class attendance
[31] Kaur, Gaganpreet, Dheerendra Singh, and Simranjeet
system." Engineering Technology and Kaur."Electrocardiogram (ECG) as a Biometric
Technopreneuship (ICE2T), 2014 4th International Characteristic: A Review." (2015).
Conference on. IEEE, 2014.
[17] Kaul, Amit, A. S. Arora, and Sushil Chauhan. "ECG
based human authentication using synthetic ECG
template." Signal Processing, Computing and Control
(ISPCC), 2012 IEEE International Conference on. IEEE,
2012.
[18] Ye, Can, Miguel Tavares Coimbra, and BVK Vijaya
Kumar. "Investigation of human identification using
two-lead electrocardiogram (ECG) signals."Biometrics:
Theory Applications and Systems (BTAS), 2010 Fourth
IEEE International Conference on. IEEE, 2010.
[19] Biel, Lena, et al. "ECG analysis: a new approach in
human identification."IEEE Transactions on
Instrumentation and Measurement 50.3 (2001): 808-
812.
[20] Matos, André Cigarro, André Lourenço, and José
Nascimento."Embedded system for individual
recognition based on ECG Biometrics." Procedia
Technology 17 (2014): 265-272.
[21] Belgacem, Noureddine, et al. "ECG based human
authentication using wavelets and random forests." Int J
Cryptography Inf Secur 2 (2012): 1-11.
[22] Singh, Yogendra Narain, and Sanjay Kumar Singh.
"Evaluation of electrocardiogram for biometric
authentication." (2011).
[23] Da Silva, Hugo Plácido, et al. "Finger ECG signal for
user authentication: Usability and
performance." Biometrics: Theory, Applications and
Systems (BTAS), 2013 IEEE Sixth International
Conference on. IEEE, 2013.
[24] Lourenço, André, Hugo Silva, and Ana Fred. "Unveiling
the biometric potential of finger-based ECG signals."
Computational intelligence and neuroscience 2011
(2011): 5.
[25] Nawal, Meenakshi, and G. N. Purohit. "Ecg based
human authentication: A review." Int. J. Emerg. Eng.
Res. Technol 2.3 (2014): 178-185.
[26] Israel, J. Irvine, A. Cheng, M. Wiederhold, and B.
Wieder- hold. ECG to identify individuals. Pattern
Recognition, 38(1):133–142, 2005.
[27] Coutinho, David Pereira, Ana LN Fred, and Mario AT
Figueiredo."One-lead ECG-based personal identification