| Popis: |
The handcrafted features extraction methods have achieved remarkable results in ECG based biometric identification. However, they are sensitive to many factors: (1) intra and inter-individual variability, (2) heart rate variability, (3) powerline interference, baseline wander and muscle artifacts. To deal with these issues, deep learning approaches have been proposed to extract automatically the important features almost from original data without any preprocessing step (i.e., The original ECG signal mostly contains noise). Unlike conventional ECG based biometric approaches, which based either on fiducial and non-fiducial methods, the proposed approach can be implemented on end to end system using deep learning approach with a blind segmentation step. The robustness of the proposed approach has been assessed on some well-known ECG datasets: The ECG-ID, MITI-BIH, CYBHI, and PTB databases. In addition, the results obtained from the proposed approach outperform other state-of the-art approaches. |
