Thermal-Polarimetric and Visible Data Collection for Face Recognition

One of the less-explored topics in automated face recognition research is exploiting the thermal IR bands along with the visible band. One such application is matching a thermal face image with visible spectrum face images for interoperability with existing biometric face databases and watch lists. One of the challenges to this cross-modal matching is the substantial differences in thermal and visible face signatures due to phenomenology. In this data collection, we explore the use of a relatively new sensor for face recognition applications, a polarimetric IR imager. When processed, the data collected by this imager provide geometric face features that result in better cross-modal recognition performance than conventional IR data. This data collection also explores the effect of several independent factors on face recognition performance: 1) subject-to-sensor distance, 2) face aspect/pose angle, and 3) facial expressions.