Abstract
In recent years, Electrocardiogram (ECG) biometric authentication has emerged as a hot topic in biometrics research due to its unique advantages including intrinsic aliveness characteristics and convenience for users. However, due to the non-stationary and nonlinear nature of ECG signals, there are still some challenges to be addressed for the application of ECG biometric authentication. In this paper, we propose a method that employs the short-time fourier transform (STFT) and a local binary descriptors learning method for ECG biometric authentication. Specifically, we first convert ECG heartbeats into two dimensional spectrogram images by STFT. Second, we extract pixel differential vectors (PDVs) from each point in the spectrogram images of the training ECG heartbeats. Third, we learn a projection matrix to map these PDVs into low-dimensional binary descriptors with three objectives: 1) The error between the original PDV and binary descriptor is minimized. 2) The intra-class variation of the local binary features is minimized and the inter-class variation of the local binary features is maximized. 3) The L2,1 norm of the learned binary descriptors is minimized. Finally, we represent each spectrogram as a histogram feature by clustering and pooling these binary descriptors. Experiments on the database verify that the proposed method outperforms other existing ECG biometric authentication methods in terms of performance.