Pattern Analysis Towards Human Verification using Photoplethysmograph Signals

Photoplethysmogram (PPG) is a biomedical signal capable of detecting blood volume changes in the microvascular bed of tissues. As PPG signals are more intimate, keen, hard to replicate, and steel, they are dedicated to providing a more secure biometric approach for user recognition. This work presents a low cost, nonintellectual pattern recognition system for biometric authentication. In the proposed methodology, raw PPG signals obtained from 20 subjects are denoised using Empirical Mode Decomposition (EMD) and reconstructed using the first three IMFs. A combination of twenty time and frequency domain features is extracted from the preprocessed PPG signals. Subsequently, a range of different classifiers, including Support Vector Machines (SVM), Decision Tree (DT), and K-Nearest Neighbors (KNN), was used to classify the features extracted. When trained to quadratic SVM classifier using 10-fold cross-validation, the system achieves an accuracy of 93.10%.