Don’t You Forget About Me: A Study on Long-Term Performance in ECG Biometrics

Abstract

The performance of biometric systems is known to decay over time, eventually rendering them ineffective. Focused on ECG-based biometrics, this work aimed to study the permanence of these signals for biometric identification in state-of-the-art methods, and measure the effect of template update on their long-term performance. Ensuring realistic testing settings, four literature methods based on autocorrelation, autoencoders, and discrete wavelet and cosine transforms, were evaluated with or without template update, using Holter signals from THEW's E-HOL 24h database. The results reveal ECG signals are unreliable for long-term biometric applications, and template update techniques offer considerable improvements over the state-of-the-art results. Nevertheless, further efforts are required to ensure long-term effectiveness in real applications.

An End-to-End Convolutional Neural Network for ECG-Based Biometric Authentication

“…However, ECG signals are greatly influenced by noise and variability [22,24], especially in off-the-person settings, which require more robust recognition approaches. Although researchers have recently started to use deep learning techniques to achieve better performance and robustness [13,27,30,38,39], current deep approaches still rely on separate predefined feature transforms and/or noise removal techniques, which are not optimized for the task at hand and therefore limit the achievable performance.…”

ELEKTRA: ELEKTRokardiomatrix application to biometric identification with convolutional neural networks

“…Then, the works found in [177], and [26] confirm an evident decay over time when training with one session and testing with a session from another day using the E-HOL-03-0202-003 and CYBHi databases, respectively. In brief, even though the use of the EKG guarantees the universality (everyone alive is beating) and permanence (cardiac signals are stable during at least five years [178]) of the system, presumably, there exists a need of training with more than one day to achieve good performance over time stability with an EKG identification method.…”

“…In fact, in a real-world scenario, to overcome the possible obstacles with permanence over the EKG signal, a good approach could be to retrain the last layer of the CNN (performing transfer learning as it is suggested in [177]) each time a user is being identified into the system to re-enrol each subject each time and adapt the model and the biometric system to possible changes over time.…”