Implementation of artificial features in improvement of biometrics based PPG

Abstract

Biometrics can provide more privacy and reliability to recognize a person, based on physiological information or bio-signals of him which is specific and inherent of that person and cannot be disguised by other individuals. Photoplethysmogram (PPG) is often considered as one of the non-invasive easy to access bio-signals of human being that is replete with information about cardiac activity, respiration, blood pressure, autonomic function, etc., thus can be used as a human ID as long as being alive is concerned. In this paper, 97 parametric features of PPG were defined for decision making and classification. Using Forward Feature Selection algorithm, 30 superior features were ranked respectively. Results of four classifiers were investigated: K-Nearest-Neighbors, Gaussian Mixture Model, Parzen Window and Fuzzy K-Nearest-Neighbors classifiers. These classifiers were applied on two groups of artificial features extracted from combination of primary features. This study shows that human authentication using PPG can be achieved with EER of 2.17% ± 0.31%.