ECG Biometric Identification Using Phase Transform and Wavelet Scattering Network

Electrocardiogram (ECG) signal is a typical time-varying nonlinear signal. The identification methods based on ECG signal has problems such as low recognition accuracy, weak model generalization ability and complex calculation on small sample data. The paper develops an efficient method using Phase Transform (Phase Transform, PT) and wavelet scattering theory (Wavelet Scattering Network, WSN). First, the ECG signal is extended to many ECG signals on PT method. These signals reveal trends of the ECG signal under different states and capture intrinsic information hidden in phase of ECG signals. and then the relevant features are extracted from them by the constructing wavelet scattering network. The effects of step size in PT, invariance Scale, qualify factors and network orders are analyzed about performance of the algorithm. Support vector machine (SVM), ensemble subspace discriminant (ESD) and K-Nearest Neighbors (KNN) as classifiers to evaluate it. The proposed method achieved identification accuracy of 99.47% for the ECG-ID database, and can achieve better feature extraction performance on small sample datasets.