Multi-scale differential feature for ECG biometrics with collective matrix factorization
Introduction
Biometrics is the field of study that models the identity of people using their physiological or behavioral traits [1], such as face [2], fingerprint [3], iris [4], finger vein [5], gait [6], and so forth. Over the past decade, some biosignals that are generally used for medical purposes, including but not limited to electrocardiogram (ECG) [7], electromyogram (EMG) [8] and electroencephalogram (EEG) [9], have been examined as biometrics for multiple reasons.
Compared with other biometric traits, ECG biometrics have several unique advantages, which can be summarized as follows. 1) ECG is recorded by attaching sensors to the body, which means that it can only be captured from a living person. 2) ECG is difficult to replicate or spoof, leading to the high security of ECG biometrics. 3) ECG signals can be acquired from all living individuals regardless of whether they are healthy; thus, ECG biometrics are universal for all people. In addition, ECG signal-capturing devices are becoming increasingly portable and inexpensive, such as wearable devices. The acquisition method makes ECG biometrics superior to several other biometric systems because acquiring other biometric signals may distract the user. 4) The acquired ECG is a one-dimensional signal that is easy to store and process.
G.E. Forsen et al. [10] first proposed using ECG as a human identification trait in 1977, and the most influential work on ECG biometrics was reported by Biel [11] in 2001. Since then, ECG biometrics have attracted increasing research interest and are regarded as one of the most promising biometric techniques. In general, the ECG signal is a quasiperiodic signal with a frequency of 1-1.5 heartbeats per second [12]. Exploiting the repetitive property can increase biometric system accuracy. As shown in Fig. 1, a healthy ECG heartbeat has six fiducial points, namely, P, Q, R, S, T, and U.
Many challenges are encountered when processing ECG biometrics. Since sensors need to be attached to the body to acquire the ECG regardless of whether the configuration is on-the-person settings (medical acquisitions) or off-the-person settings, the interface between the body and electrodes may create contact noise, which affects the quality of the ECG signal to some degree. In addition, ECG signals are easily affected by physiological and psychological changes. For example, it could change along with activities, diets, diseases, positions of the electrodes, and other factors [7]. With these considerations, it is urgent to propose proper feature extraction methods or other techniques to overcome these challenges. Thus, in this paper, an effective feature extraction framework for ECG biometrics is introduced.
We design a novel feature extraction framework to generate a discriminative ECG biometrics feature representation, which can help improve the ECG biometrics accuracy. First, the ECG signal is represented by a new type of base feature. Inspired by one-dimensional Multi-Resolution Local Binary Patterns (1DMRLBP for short) [12], which was proposed by Louis et al., we design an improved multi-scale differential feature fused of one dimensional multiresolution local binary patterns (MSDF-1DMRLBP for short) by fusing the Multi-Scale Differential Feature (MSDF) with 1DMRLBP to act as the intermediate base feature of the ECG signal. The proposed base feature takes the differential changes with temporal variations into consideration and improves the capability of capturing micro texture and multi-scale differential signal characteristics. Then, we extract discriminative features from the output of the former step and generate the final enhanced features. Specifically, we adopt the collective matrix factorization (CMF) technique, which simultaneously embeds MSDF-1DMRLBP and label information, to seek a latent implicit feature space. The merits of this step are twofold: 1) The final features could preserve the intra-subject and inter-subject similarities. 2) The final features are robust and applicable to the noisy 1D ECG signals.
The contributions of this paper are summarized as follows:
- •
First, we propose a novel multi-scale differential feature extraction method to efficiently capture micro texture and multi-scale differential signal characteristics.
- •
Second, we generate the final feature by leveraging the CMF technique, which simultaneously embeds MSDF-1DMRLBP and label information. Due to CMF and semantic embedding, we can make full use of the supervised information and maximally preserve the intra-subject and inter-subject similarities.
- •
Extensive experiments are conducted on four ECG datasets. The results demonstrate that the proposed method outperforms several state-of-the-art methods.
The remainder of this paper is organized as follows. Section 2 reviews the literature on ECG biometrics. Section 3 describes the proposed method in detail. Section 4 reports the experimental results and provides a comprehensive analysis. Finally, Section 5 presents the conclusions and future work.
Section snippets
Studies on ECG biometrics
Based on the ECG fiducial points in Fig. 1, research on ECG biometrics can be divided into two main categories, i.e., fiducial-based and non-fiducial-based approaches.
In fiducial-point-based approaches, features are extracted from fiducial points and distances between points, angles, areas, amplitudes, and so forth. There are some related examples, i.e., Juan et al. [13] extracted the time intervals and difference values in amplitude among several fiducial points, such as PR, QR, SR and TR. L.
Proposed method
To perform ECG biometrics, we designed a novel ECG biometrics framework. First (mentioned in Section 3.2), an improved base feature extraction algorithm is introduced to generate the intermediate base feature, i.e., MSDF-1DMRLBP. Second (mentioned in Section 3.3), CMF is leveraged to generate the final discriminative feature for ECG signals. Finally, the matching procedure is performed for the testing data. Before presenting the framework, we first preprocess and segment the ECG signals into
Database
To comprehensively validate the effectiveness of the proposed ECG-based feature extraction method, we conducted extensive experiments on four widely used benchmark datasets [42], i.e., MIT-BIH Arrhythmia (MITDB)[43], ECG-ID [44], PTB Diagnostic ECG Database [45] and the University of Toronto ECG Database (UofTDB) [46]. The summary of the four adopted datasets is presented in Table 1.
ECG-ID is a database entirely focused on biometrics. Twenty-second ECG recordings are collected from 90 subjects
Conclusion andfuture work
In this paper, we propose a novel ECG biometrics framework, including Multi-Scale Differential Feature fused with one-dimensional Multi-Resolution Local Binary Patterns (MSDF-1DMRLBP) and final feature learning with collective matrix factorization (CMF). First, the MSDF-1DMRLBP is extracted, which can express the intrinsic characteristics of the ECG, such as the local wave shape and the growth direction of the ECG signal, preserving the ECG heartbeat morphology about micro texture and
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgements
This work was supported in part by the National Natural Science Foundation of China under Grants U1903127, 61703235 and 61876098 and in part by the Key Technology Research and Development Program of Shandong Province under Grant 2018GGX101032, 2019GGX101056.
Kuikui Wang received her master degree in computer science from the Shandong University, in 2014. Currently, she is pursuing her Ph.D. degree at Shandong University. Her main research interests are biometrics and machine learning.
References (50)
Human recognition using fisher discriminant analysis of heartbeat interval features and ecg morphology
Neurocomputing
(2015)- et al.
Verification of humans using the electrocardiogram
Pattern Recognit. Lett.
(2007) - et al.
Human identification by quantifying similarity and dissimilarity in electrocardiogram phase space
Pattern Recognit.
(2009) - et al.
Ecg biometric authentication based on non-fiducial approach using kernel methods
Digit. Signal Process.
(2016) - et al.
Ecg-based personal recognition using a convolutional neural network
Pattern Recognit. Lett.
(2019) - et al.
Fusion of electrocardiogram with unobtrusive biometrics: an efficient individual authentication system
Pattern Recognit. Lett.
(2012) - et al.
Multimodal biometric system combining ECG and sound signals
Pattern Recognit. Lett.
(2014) - et al.
Scratch: a scalable discrete matrix factorization hashing for cross-modal retrieval
Proceedings of the International Conference on Multimedia
(2018) Biometric liveness detection
Inf. Secur. Bull.
(2005)- et al.
FaceNet: a unified embedding for face recognition and clustering
Proceedings of the Conference on Computer Vision and Pattern Recognition
(2015)
Automated latent fingerprint recognition
IEEE Trans. Pattern Anal. Mach. Intell.
Deepirisnet: deep iris representation with applications in iris recognition and cross-sensor iris recognition
Proceedings of the International Conference on Image Processing
Finger vein recognition with anatomy structure analysis
IEEE Trans. Circuits Syst. Video Technol.
Individual recognition using gait energy image
IEEE Trans. Pattern Anal. Mach. Intell.
Ecg biometric recognition: a comparative analysis
IEEE Trans. Inf. For. Secur.
Electromyography analysis for person identification
Int. J. Biomet. Bioinf.
EEG/ECG signal fusion aimed at biometric recognition
Proceedings of the International Conference on Image Analysis and Processing
Personal Attributes Authentication Techniques.
Technical Report
Ecg analysis: a new approach in human identification
IEEE Trans. Instrum. Meas.
Continuous authentication using one-dimensional multi-resolution local binary patterns (1DMRLBP) in ECG biometrics
IEEE Trans. Inf. For. Secur.
ECG authentication for mobile devices
IEEE Trans. Instrum. Meas.
Robust ECG biometrics by fusing temporal and cepstral information
Proceedings of the International Conference on Pattern Recognition
Ecg biometrics: a robust short-time frequency analysis
Proceedings of the International Workshop on Information Forensics and Security
ECG biometric identification for general population using multiresolution analysis of DWT based features
Proceedings of the International Conference on Information Security and Cyber Forensics
Electrocardiogram (ECG) biometric authentication using pulse active ratio (par)
IEEE Trans. Inf. For. Secur.
Cited by (32)
BAED: A secured biometric authentication system using ECG signal based on deep learning techniques
2022, Biocybernetics and Biomedical EngineeringCitation Excerpt :The method extracts fiducial features of each heartbeat to compute statistical features and decisions made with sequential procedures. K.Wang et al. [31] extracted QRS Complex, onset and offset of P, Q, S and T waves for Lead-I ECG and performed biometric recognition. Methods based on non-fiducial features do not recommend characteristic points for extracting feature sets.
Robust multi-feature collective non-negative matrix factorization for ECG biometrics
2022, Pattern RecognitionCitation Excerpt :Hence a challenging problem for LBP-based recognition methods is to design a robust method to extract the latent semantics. The widely used collective matrix factorization (CMF) can extract the latent semantics and remove redundant information [15–17]. Most CMF-based methods assume that all input data matrices share the same factor and should have the same semantics.
Multi-view discriminant analysis with sample diversity for ECG biometric recognition
2021, Pattern Recognition LettersECG-based biometric under different psychological stress states
2021, Computer Methods and Programs in BiomedicineCitation Excerpt :Different people show different ECG features, but it is difficult to design representative manual features. Kuikui Wang et al. proposed a Multi-Scale Differential Feature for ECG biometric identification with Collective Matrix Factorization (CMF) in 2020 [4]. By extracting multi-scale differential features from the one-dimensional ECG signal, and at the same time extracting discriminative information from the intermediate basic features, a robust ECG feature is finally generated.
Kuikui Wang received her master degree in computer science from the Shandong University, in 2014. Currently, she is pursuing her Ph.D. degree at Shandong University. Her main research interests are biometrics and machine learning.
Gongping Yang received the Ph.D. degree in computer software and theory from Shandong University, China, in 2007. He is a Professor with the School of Software, Shandong University and an adjunct professor in the school of Computer, Heze University. His research interests are pattern recognition, image processing, biometrics, and so forth.
Yuwen Huang received the master degree in computer science from Guangxi Normal University, in 2009. He is an associate professor in the school of computer, Heze university and pursuing his Ph.D. degree at Shandong University. His research interests include ECG recognition, biometrics and machine learning.
Yilong Yin received the Ph.D. degree from Jilin University, Changchun, China, in 2000. From 2000 to 2002, he was a Post-Doctoral Fellow with the Department of Electronic Science and Engineering, Nanjing University, Nanjing, China. He is currently the Director of the Machine Learning and Applications Group and a Professor with Shandong University, Jinan, China. His research interests include machine learning, data mining, and biometric.