Fig. 4: Decision boundary and KDE of rPPG signals classified by a linear SVM.

This figure showcases the decision boundary determined by a linear support vector machine (SVM) in the classification of remote photoplethysmography (rPPG) signals, alongside the kernel density estimation (KDE) for signals from different classes. The visualization elucidates the separation achieved by the linear SVM and provides a density-based perspective on the distribution of rPPG signals across the classified groups. We considered labeled signals reconstructed through the CHROM, GREEN, and OMIT rPPG methods from the PURE and LGIPPGI datasets combined with a window size of 5 s. For each comparison, we considered the best SQI (see Tables 1 and 2): a N_SQI for Excellent signals vs. Acceptable signals; b K_SQI for Unfit signals vs. Acceptable signals; and c N_SQI for Excellent signals vs. Unfit signals.