Exploiting polarization-state information for cross-spectrum face recognition

Face recognition research has primarily focused on the visible spectrum, due to the prevalence and low cost of visible cameras. However, face recognition in the visible spectrum is sensitive to illumination variations, and is infeasible in low-light or nighttime settings. In contrast, thermal imaging acquires naturally emitted radiation from facial skin tissue, and is therefore ideal for nighttime surveillance and intelligence gathering operations. However, conventional thermal face imagery lacks textural and geometrics details that are present in visible spectrum face signatures. In this work, we further explore the impact of polarimetric imaging in the LWIR spectrum for face recognition. Polarization-state information provides textural and geometric facial details unavailable with conventional thermal imaging. Since the frequency content of the conventional thermal, polarimetric thermal, and visible images is quite different, we propose a spatial correlation based procedure to optimize the filtering of polarimetric thermal and visible face images to further facilitate cross-spectrum face recognition. Additionally, we use a more extensive gallery database to more robustly demonstrate an improvement in the performance of cross-spectrum face recognition using polarimetric thermal imaging.