Enhancing Generalized Electrocardiogram Biometrics Transformer

Using electrocardiogram (ECG) as biometrics has been explored over the years because it fits well in health monitoring applications. Most of the ECG biometrics models are specialized and can only work with very specific conditions otherwise fine-tuning, retraining or redesigning are required. Generalized ECG biometrics models are more suitable for real world applications but require large and diverse datasets to train. In this study, we introduced three databases into training the generalized ECG biometrics transformer to enhance its generalization capability. The model scored 2.93, 0.86, 2.47 and 0.29% in equal error rate for authentication task and 97.15, 99.87, 97.46 and 100.00% in identification accuracies on AFDB, NSRDB, STDB and CEBSDB respectively.