Biometric identification based on electrocardiogram Using Markov Transition Field and Hybrid Network

Identification based on electrocardiogram (ECG) has attracted attention of researchers. One-dimension features of ECG signal are extracted and classified is effective in the rest state; however, the scheme performance in activity state is unsatisfactory. Because ECG waveform is influenced by the subject's emotional and physical states. To address ECG recognition during activity, an approach for personal identification of ECG, incorporating the first-order Markov chain and phase transform is proposed. The proposed approach involves the segmentation of ECG into sub-signals, each comprising multiple heartbeats, carrying rich feature information. Sub-signals are converted into images using the Markov Transition Field (MTF) Additionally, the Phase Transform (PT) algorithm is employed to unveil inter-beat information within ECG signals. For the generated images using the MTF method, a hybrid network, combining the wavelet scattering network (WSN) and Resnet, is employed for feature extraction and classification. The impact of sub-signal lengths and the utilization of the PT algorithm in capturing inter-class information on the recognition model's performance are discussed in experiments. The proposed method's feasibility is substantiated validation on two public datasets: ECG-ID and ECG-GUDB. In the ECG-ID dataset, the recognition system attains a commendable accuracy of 90.26%. In the ECG-GUDB dataset, assessments under various conditions reveal impressive performance, achieving 94.34% accuracy in the resting states, 93.17% in the moving states, and 92.73% accuracy in combined moving and resting states. These results surpass the performance of the majority of existing algorithms.