Covert and robust facial recognition of individuals is a goal for military check points, law enforcement, and border control. Facial recognition in the visible spectrum is an important biometric modality. The performance of facial recognition systems can significantly degrade in uncontrolled settings. This in turn hinders database enrollment and limits the circumstances under which verification and identification can take place. While visible imagery has increased in maturity for facial recognition, there are situations in which visible will not work and thermal imaging is required, particularly in dark settings. However, thermal imaging relies on temperature differences to provide contrast in scenes, but for scenes that don’t have significant temperature differences, recognition or identification cannot be accomplished. Thermal signatures of human faces rarely have sufficient contrast to provide the features necessary for recognition algorithms.
Polarization enhanced thermal cameras have been shown to produce significant contrast on human faces, providing rich geometric and textural features similar enough to visible signatures that there is good potential identification of facial features. Further, this approach enables passive (i.e. no external illumination required) day and night capability. Polarization can address the capability gaps in low-light and nighttime by acquiring naturally emitted radiation in the thermal infrared spectrum from facial skin tissue. The polarimetric images capture key facial details and geometry not available in conventional thermal imagery. Facial recognition software uses novel algorithms and is capable of matching polarimetric thermal facial imagery to the visible spectrum face. This approach provides interoperability with existing biometric databases containing visible-only face imagery (e.g., watchlists).
- Infrared sensor that delivers daylight detail at night.
- Distinguishing details surpass traditional thermal imagery.
- New technology enables recognition and decision making due to improved visibility.
By adding the polarization sensing capability to existing thermal sensors, contrast can be enhanced over that of conventional thermal imaging due to differential polarization emission without external lighting. Figure 1 shows an example of how a thermal image (center) loses the detail present in the visible imager (left). The thermal polarimetric image (right) restores the detail of the visible image due to polarimetric, and not thermal, differences. The level of detail in the polarimetric image may be sufficient to compare to archived visible face data for identification.
Infrared systems enable imaging of scenes using the inherent thermal emission of the objects in the scene, thus enabling effective imaging in the dark. Thermal imaging relies on temperature differences to provide contrast in scenes, and for scene elements that don’t have significant temperature differences; recognition or identification can be difficult. Thermal polarimetric sensing does not rely on external illumination, so this approach will work well in situations where infrastructure resources are limited or restricted. Thermal polarimetric face data can be matched to visible databases of facial signatures (i.e. thermal polarimetric-to-visible face recognition). This will reduce or eliminate the need to collect face data with the new sensor, offering interoperability with existing watch lists and databases. A thermal polarimetric imager can be used for more robust facial recognition systems for biometric identification, under varied (or no) illumination conditions in a small, portable package.