Practical human authentication method based on piecewise corrected Electrocardiogram

In this paper, we present a practical Electrocardiogram-based (ECG) human authentication method which is very suitable for deployment in mobile devices like smart phones. A user only needs 20 seconds to register his ECG template at the first time and about 4 seconds to authenticate afterwards. We firstly extracted heartbeats of every cardiac cycle and then separately corrected PQ segments, QRS segments and JT segments divided by fiducials detected earlier of every heartbeats. At last, we reduced the feature dimension to 80 by discrete cosine transformation. 85 healthy subjects' ECG records were selected from 3 widely used public ECG databases. We adopted template matching method as classifier in authentication experiment and the equal error rate was 1.57%. We also tested the classification performance of the ECG features by using neural network and SVM classifier separately. The accuracy rate was 96.6% and 97.6% which fully demonstrated the prospect of using the features in identification applications too.