Subject Authentication using Textural Features of Recurrence Plot

Increasing dependence on online transactions combined with continuing developments in information systems have led to a demand for accurate, reliable subject identity authentication systems. Various biometric-based technologies for identify authentication have already been proposed, including face, retina, and fingerprint recognition. However, this study, proposes an authentication system based on textural features extracted from the recurrence plot (RP) of three physiological signals:Photoplethysmography(PPG), Electrocardiogram (ECG) and Capnograms (CO$_{\mathbf{2}}$). The signals are divided into segments from which a recurrence plot is generated. From there, textural features are extracted to form feature vectors subsequently fed into classifiers for user identity authentication. These classifiers are random forest (RF), naïve Bayes (NB), linear discriminant analysis (LDA), and support vector machine (SVM). This work also investigates the fusing feature vectors of different physiological signals into a single feature vector. The most promising results using feature vectors of individual signal: PPG, ECG, and CO$_{\mathbf{2}}$, were obtained using RF classifier with segment lengths of 5.5 and 18 seconds; this approach achieved an average accuracy of 98.74%, 98.36%, and 98.41%, respectively. The fused features of the three signals also provided very promising results using an RF classifier, with an average accuracy of 99% and 99.31% using segment lengths of 2 and 15 seconds.