Study on Matthew Effect Based Feature Extraction for ECG Biometric.

Electrocardiogram (ECG) is a “live” signal and is very difficult to be copied or forged, which make it becoming a competitive biology material for biometric. But, due to ECG signals are easily be affected by the external environment and human states (physiological or psychological), therefore, finding stable features becomes one of the key issues in the research. The paper proposed a feature extraction method based on ECG superposition matrix of single heartbeat ECG. By matrix segmentation and similarity comparison, ECG stable feature distribution area can be selected, and stable feature sets are constructed. And through further study, it is found that a large number of heartbeats are need to build superposition matrix. For solving the problem, the paper proposed a Matthew effect based method, by which superposition matrix can be constructed by only 10 heartbeats. Experiments results showed that average TPR of 100 heartbeats was reaching 83.21, 83.93 and 80% respectively. And that of 10 heartbeats with Matthew Effect reach 80.36, 82.68 and 80.77% respectively, which is competitive compare to 100 heartbeats superposition matrix.