Missed isochromatic cardiac pulsation in remote photoplethysmography: detection, impact, and removal

Remote photoplethysmography (rPPG) extracts the cardiac pulse from facial video by isolating chromatic variation due to wavelength-dependent hemoglobin absorption. We report that this signal model is incomplete. A second cardiac-frequency signal is present in skin video: an isochromatic pulsation along the skin's mean reflectance direction, arising from ballistocardiographic modulation of skin surface geometry and changes in the effective scattering cross-section of the dermal vascular bed. This isochromatic component is conflated with the chromatic signal by every rPPG method we evaluated, including chrominance-based methods (POS, CHROM) and other projection or decomposition methods (OMIT, LGI, PBV, ICA). The two signals share the cardiac frequency but differ in phase and signal fidelity, and their relative directions vary with skin tone, lighting, and recording conditions. To assess whether the isochromatic component contributes significant error to rPPG, we constructed chrominance Phase-Aware Cardiac Eigenprojection (cPACE), a method that removes the isochromatic direction via a subject-adaptive projection orthogonal to the measured skin reflectance before extracting the cardiac signal through eigen-vector projection. Across UBFC-Phys (rest and speech) and the CMU-rPPG India and Sierra Leone cohorts, cross-ROI phase-locking values of the recovered signal are systematically higher under cPACE than under any other method evaluated, indicating cleaner waveform recovery. The resulting heart rate estimates achieve mean absolute error of 1.7, 3.7, 1.7, and 1.1 BPM, respectively, on these cohorts, compared to 2.8, 9.5, 16.8, and 17.8 BPM for POS with homodyne envelope correction applied for parity. Together, these results show that the isochromatic component is a major source of error in existing rPPG methods, and future methods should explicitly account for it.