Biometric recognition is widely integrated in many
types of systems. Fingerprint, face or voice are some of the best well-known modalities.
Biometric recognition was traditionally used in link to identify subjects
within security contexts. One desired advantage in security applications is the
use of biometric signals strongly to impersonation. Electrocardiogram (ECG) bio
signal may be considered as one this kind. The ECG signal is widely used for
medical purposes like pathology detection and monitoring patient. The evolution
of the capture devices allow to use this signal in biometric recognition.
ECG is a record of time varying bio-potential
generated by electrical activity of the heart.
Authentication using ECG signal often consists of
four main steps: 1) pre-processing for noise reduction and artifacts removal,
2) feature extraction, 3) classi?cation.
The previous works for ECG based Biometric
recognition mainly differ in feature extraction techniques and classifications.
The feature extraction techniques are divided in mainly two categories,
fiducial and non-fiducial methods. The first one, are the
points of reference of the signals such as distance between points, amplitude
of one or more points or area of the waves. The accuracy of the result is strongly
associated with feature recognition method. To solve this disadvantage some
non-fiducial methods are performed.
The most famous public databases used in the
literature are: PTB DB, MITDB, MITSUP, NSRDB, ECB, QT DB, LTSTDB and ECG ID,
all available at physionet.org.
Non-fiducial option consists in to obtain
discriminative coefficients of the signal such as Discrete Cosine Transform
(DCT) coefficients, covariance matrix, wavelet coefficients, daubechies wavelet
etc. Some of those are based in the premise that one heartbeat is very similar
than other heartbeat of the same person.
Venkatesh N and Srinivasan Jayaraman in 2010
presented a paper in an IEEE conference; it proposes a novel approach for
person’s identification and authentication based on single lead ECG. They used a two stage classification
techniques. In two stage classifications FLDA was used with k-NNC at the first
stage followed by DTW classifier at the second stage which yielded 100%
recognition accuracy and QRS based threshold feature technique for
authentication is used. For filtering band pass filter was used followed by Tompkins
method and zero crossing for feature extraction. ECG
classification had been performed using Dynamic Time Warping (DTW) and Fisher’s
Linear Discriminant analysis (FLDA) with Nearest Neighbor Classifier (NNC) for
single stage classification. Further FLDA and DTW were combined to yield better
results. For verification an equation was used, if the QRS interval difference
between the test user trial and the identified user trial exceeds
experimentally found threshold ‘?’ then he was a valid user.
Takoua Hamdi et al. in 2014, proposed to use slope
extraction and angle extraction algorithms for feature extraction from ECG signal.
For classification they used multilayer neural networks (Multi-Layer Perception
MLP).It is capable of modelling complex structures and noisy irregular data.
They divide their experiment into to setup learning based and test based by
applying segmentation for sampling. The recognition rate was found to be
Shyan -Lung Ling et al. in 2014, presented chaotic
theory for specific biological characteristics extraction which confirms using
the ECG signal for person recognition even during exercising. During the rest
state ECG signal was converted in phase plane using reconstruction and chaos extractor
is applied to capture major indices of a chaotic ECG signal. These parameters
were used for dynamic ECG also. The ECG data is analysed using root mean square
value, non-linear Lyapunov exponent and correlation dimension. SVM (support
vector machine) was used to classify and identify the best combination and
kernel function of SVM. It was found that the accuracy was identified to be
much higher than the other techniques applied for identification.
Ahmad Ba-Hammam et al. in 2017 proposed to use
Linear Discrimination Analysis (LDA) at classification stage and also develop a
parallel processing technique if a new subject would have been added. They
extracted eight features from ECG. LDA maximizes the Fisher score function.
They shows that their algorithm with achieve a better performance in comparison
to SVM technique used earlier for classification as LDA is simple to implement
Ronald Salloum and C.-C. Jay Kuo in 2017 proposed
the use of recurrent neural networks (RNNs) to develop an effective solution to
two problems in electrocardiogram (ECG)-based biometrics:
identi?cation/classi?cation and authentication. The data was obtained from MIT
–BIH database and send for pre-processing and segmentation. R-peaks were
detected using Pan-Tompkins algorithm. LSTM based RNN was performed for
identification and authentication and accuracy was determined by equal error
rate, false acceptance rate and false rejection rate. It was found that the
classification accuracy was achieved to 100% .This technique was found to be
equally suitable for arrhythmia patients.
So, currently neural networks techniques for
identification and classification of ECG signals were used increasingly in ECG
based Biometric systems.