Influence of ROI Selection for Remote Photoplethysmography with Singular Spectrum Analysis

Heart rate (HR) extraction using video-based remote photoplethysmography (rPPG) technology has been one of the hot topics in non-contact physiological parameter detection in recent years. The selection of regions of interest (ROI) is critical to the quality of rPPG measurements, especially for remote HR measurement under uneven illuminations. In this paper, we investigate the effect of ROI selection for rPPG using the singular spectrum analysis (SSA). Specifically, we define patches from a full ROI covering skins, and the patches are ranked and selected through quality metrics. The impact of ROI on the performance of SSA method is then evaluated through using different numbers of optimal patch ROIs. We demonstrate the effectiveness of the proposed method on the public dataset COHFACE. It is experimentally given that the selection of the optimal patch ROIs can most effectively eliminate illumination noise and achieve reliable HR measurements.