A novel unified framework for noise-robust ECG-based biometric authentication

This paper presents a simple unified ensemble averaged heartbeat extraction framework for noise-robust ECG-based biometric authentication using heartbeat morphology. The proposed method consists of three major steps: preprocessing, ensemble averaged beat construction, and similarity matching. At the preprocessing stage, the signal blocking and mean removal operations are performed. The ensemble averaging stage includes the steps of: discrete cosine transform (DCT) based filter for simultaneous removal of BW and PLI noises, straightforward Gaussian derivative filter (GDF)-based R-peak detector, period normalization and peak-centering, and ensemble averaging computation. At the similarity matching stage, we study the performance of both time-domain and wavelet-domain distance metrics for finding the similarity between a test ensemble ECG beat template and an enrolled ECG beat template. The performance of the proposed framework is tested and validated using different types of ECG signals taken from the standard ECG databases. Results show that the ECG-biometric method using normalized cross correlation (NCC) metric provides consistent verification results as compared to the other four methods under different noisy conditions and sampling rates. For predefined threshold of 0.97, the NCC-based ECG-biometric method achieves an average false rejection rate (FRR) of 11.6% and false acceptance rate (FAR) of 5.8% for 3969 imposters and 19404 test segments from enrolled subjects, respectively.