Cardioid graph based ECG biometric using compressed QRS complex

In this paper, a Cardioid graph based feature extraction technique is applied to perform compressed Electrocardiogram (ECG) biometric at different physiological conditions. To the best of our knowledge, Cardioid graph based method has not been implemented on compressed ECG before. Another merit of this methodology is that no decompression of the compressed ECG signal is necessary before the recognition step. The QRS complexes obtained from the ECG signal is compressed using Discrete Wavelet Transform (DWT), followed by the Cardioid graph retrieval procedure. Compression is performed in three decomposition levels and with the first three Daubechies wavelets. Classification is conducted on all the three levels using Multilayer Perceptron (MLP) Neural Network. Maximum compression of 88.3% is achieved with an accuracy rate of 93.06%. For compression rate of 85%, the identification rate obtained is 95.3%. Highest recognition rate of 96.4% is attained when the compression ratio is 75%. The classification accuracy rates suggest that compressed ECG biometric in varying physiological conditions with Cardioid graph based feature extraction is feasible and is capable of producing a robust biometric system.