An approach for human identification based on time and frequency domain features extracted from ECG signals

This paper presents an effective algorithm for human identification using time and frequency domain features extracted from electrocardiogram (ECG) signal. Instead of directly using the ECG data of a person as the feature, first, it is shown that a few number of reflection coefficients extracted from the autocorrelation function of the data can efficiently perform the recognition task. In frequency domain, discrete cosine transform (DCT) based feature is employed, which offers even a better recognition performance. It is found that the proposed time and frequency domain features demonstrate a high within class compactness and between class separability. Prior to feature extraction, a pre-processing scheme is incorporated to reduce the effect of different noises and artifacts. For the purpose of recognition, a linear discriminant based classifier is employed, where the two features are jointly utilized. The proposed human identification scheme has been tested on standard ECG databases and high recognition accuracy is achieved with an extremely low feature dimension.