Design of a Hamming-distance classifier for ECG biometrics

In existing ECG-based biometric recognition systems, the feature extraction and matching are performed in Euclidean spaces. However, there are many scenarios (e.g., biometric template encryption for privacy protection, or low-complexity classification in an identification mode of operation) in which it is useful to binarize the feature vectors. The main contribution of this paper is a Hamming-distance classifier for ECG biometrics based on SPEC-Hashing. The proposed system was evaluated over a database of ECG signals from 52 different subjects that were collected at the Biometrics Security Laboratory of the University of Toronto. The EER of the Hamming-distance classifier was found to be 5.5% for closed-set matching and 14.82% for open set matching.