248 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 57, NO. 2, FEBRUARY 2008 Wavelet Distance Measure for Person Identification Using Electrocardiograms Adrian D. C. Chan, Senior Member, IEEE, Mohyeldin M. Hamdy, Student Member, IEEE, Armin Badre, Student Member, IEEE, and Vesal Badee, Member, IEEE Abstract—In this paper, the authors present an evaluation of a new biometric based on electrocardiogram (ECG) wave- forms. ECG data were collected from 50 subjects during three data-recording sessions on different days using a simple user interface, where subjects held two electrodes on the pads of their thumbs using their thumb and index fingers. Data from session 1 were used to establish an enrolled database, and data from the remaining two sessions were used as test cases. Classification was performed using three different quantitative measures: percent residual difference, correlation coefficient, and a novel distance measure based on wavelet transform. The wavelet distance mea- sure has a classification accuracy of 89%, outperforming the other methods by nearly 10%. This ECG person-identification modality would be a useful supplement for conventional biometrics, such as fingerprint and palm recognition systems. Index Terms—Biometric, electrocardiogram (ECG), intra subject variability, person identification, wavelets. I. I NTRODUCTION P ERSON identification and verification can be useful in a variety of applications. Biometrics provide an automatic method of recognizing a person based upon intrinsic physical or behavioral features, such as face, voice, retina, gait, and finger- prints. In recent years, the electrocardiogram (ECG) has been proposed as a novel biometric for person identification [1]–[4]. The ECG is an electrical signal that is associated with the con- traction of the heart (Fig. 1). A typical normal ECG pulse from a single heartbeat consists of a P wave, which is associated with the contraction of the atria; a QRS complex, which is associated with the contraction of the ventricles; and a T wave, which is associated with the repolarization of the ventricles. Whereas results in previous studies in ECG person identification demon- strate a good potential of this biometric, reproducibility of the ECG data was not assessed over sufficiently long periods (data were collected within a single measurement session). There are a number of factors that can influence ECG record- ings, including body habitus (e.g., obesity), sex, and age [5], [6]. These factors result in a large interindividual variability in ECG recordings, which causes difficulties in discerning Manuscript received December 2, 2005; revised September 13, 2007. This work was supported by the Ontario Research Network for Electronic Commerce. A. D. C. Chan, M. M. Hamdy, and A. Badre are with the Department of System and Computer Engineering, Carleton University, Ottawa, ON K1S 5B6, Canada (e-mail: adcchan@sce.carleton.ca). V. Badee was with the Department of System and Computer Engineering, Carleton University, Ottawa, ON K1S 5B6, Canada. He is now with Biosign Technologies Inc., Toronto, ON M2N 6S6, Canada. Digital Object Identifier 10.1109/TIM.2007.909996 pathologies [7]; however, this interindividual variability is ac- tually beneficial for the application of person identification. Slow changing factors, such as body habitus and age, could introduce long-term intrasubject variability, which could limit the applicability of ECG person identification or require some dynamic compensation for this variability. However, other in- fluencing factors, such as electrode placement [8] and phar- maceutical drugs [9], may be more detrimental as they could increase the intrasubject variability, even in the short term, limiting system performance. In [10], it is suggested that heart-rate variability could also be another measure useful in person identification. ECG data were recorded under seven physiological states (initial base- line, meditation, reading task, arithmetic task, initial recovery, driving task, and final recovery). To distinguish subjects, the authors utilized features including the mean and variance of the R–R interval (related to heart-rate variability), along with other ECG pulse parameters (i.e., width of the P, R, and T waves). It is likely that the discerning features in this paper are those related to the ECG pulse shape rather than the heart- rate variability; however, the independent contribution of the heart-rate-variability parameters cannot be determined from the reported results. Although these results are only for a five- person subject pool, it does provide some evidence of stability in the ECG pulse shape across different physiological states. This modality of using ECG data in person identification offers some unique advantages. It is suggested that ECG person identification would be particularly effective in health care applications as the ECG is frequently used to monitor a pa- tient’s condition [2], [4]; therefore, without any additional data requirements, this could be used to verify a patient’s identity in medical records or prior to drug administration or other medical procedures. In [11], ECG data were used as a second source of biometric data to supplement facial image data. Supplementing other biometrics with ECG information can be accomplished with little user-perceived change in the interface. For example, the ECG data could be recorded simultaneously in fingerprint and palm recognition systems. Multimodal person identification techniques can provide significant increases in performance as monomodal methods begin to saturate in their performance [12]. In the least, ECG information would provide a method for “liveness” detection, increasing system reliability [13]. In this paper, we investigate the influence of intrasubject variability by collecting data from multiple subjects across mul- tiple sessions, which, unlike previous studies, are on different days. Also dissimilar to previous studies, data collection was 0018-9456/$25.00 © 2008 IEEE