Application of Time-Frequency Analysis to Feature Extraction for Biometric Authentication with High-Frequency Electrocardiogram

Abstract

In late years, biometrics have been used as the robust security. We have focused on high-frequency electrocardiogram (HFECG) as a feature for biometric authentication. In the previous study, sampled HFECG has been applied directly to neural networks for identification, however, provided information was redundant. It have tried the reduction of information amount and the extraction of personal information from HFECG.As the extraction technique, we used matching pursuit (MP) and discrete wavelet transform (DWT) as feature extraction techniques. Identification with extracted features was performed by neural network. In previous study, 200 samples of HFECG waveform gives 100% recognition rate with 15 subjects. In MP evaluation, 95% of recognition rate was obtained with 11 samples. In DWT evaluation, 99% of recognition rate was obtained with 64 samples. MP can largely reduce redundancy and DWT can keep identification rate with reduced number of feature samples.