Shared Weighted Continuous Wavelet Capsule Network for Electrocardiogram Biometric Identification

In recent times, researchers are showing more interest in the subject of biometric identification, which uses biological traits to confirm a user's identity. Traditional authentication techniques are prone to damage, fraud, and negligence. We investigate a unique biometric based on electrocardiogram (ECG) signals generated from the heart as a biometric security attribute for access control verification. In this research, we propose a shared weighted continuous wavelet capsule network for ECG biometric identification, in which a continuous wavelet transform (CWT) is utilized to convert one-dimensional time-domain ECG signals into scalograms of two-dimensional images to obtain good quality training data. Then, a siamese capsule network framework is utilized to predict the right match or mismatch of ECG query samples using the extracted specific attributes from the scalograms. The dataset utilized in this work is collected from the Physionet MIT-BIH Normal Sinus Rhythm database. Experimental result shows that the proposed approach properly predicted ECG query samples with 99.2% accuracy, which makes our model more robust.