Single lead Electrocardiogram feature extraction for the human verification

Over the past years, Electrocardiogram (ECG) as a biometric characteristic, has been investigated in several works. The human heart is physiologically a liveness indicator. Feasibility of continuous signal acquisition and demonstration of subject aliveness, are the most important properties of ECG based authentication systems which makes them different from common authentication methods like fingerprints. In this paper, after signal denoising, two different feature extraction methods are proposed. By selecting reference beats, four artificial features are generated for every extracted feature and then they are classified using five different classifiers. As it is worthwhile to have a verification system with low number of features, the proposed method achieved to %99.38±0.04 accuracy and %0.62±0.04 EER with 5 features and SVM classifier.