Biometric template extraction from a heartbeat signal captured from fingers

Authentication of remote users is very important for security and privacy of multimedia content that is often accessed online. Heartbeat signal has emerged as biometric modality suitable for remote authentication for its privacy and liveness property. In order to improve acceptability of this modality, we propose a method of biometric template extraction from a heartbeat signal captured from fingers. An unsupervised outlier detection method is employed to select the most regular heartbeats from a specimen. In order to mitigate the effect of heart rate variability (HRV), morphology of the selected heartbeat is aligned by a piecewise-uniform method. Then, a template is formed by averaging aligned heartbeats and presented in a lower dimensional space by the principal component analysis. Authentication performance of the template was evaluated using a database collected from 112 individuals in multiple sessions by a handheld ECG device. We also implemented four state-of-the-art templates and tested them by the same database. Experimental results indicate that authentication performance of the proposed template is superior to those of any of these templates. It is also suitable for remote authentication using mobile computing devices for its computational efficiency and compactness.