Polarimetric Thermal Imaging

Passive infrared (IR) imagers, using intensity contrast for target detection, are often limited by low target-background contrast. Detecting stationary targets against cluttered backgrounds presents an even bigger challenge. Polarized signatures can be used as an additional discriminator, to improve target detection probability and reduce false alarm rate. In this research, a polarimetric thermal imager, operating in the mid wave infrared (3-5 μm), was set up using the Merlin InSb camera with three internal wire grid polarizers. Non uniformity correction and radiometric calibration were performed to compensate for differences in detector response and polarizer’s transmittance. The scene consisted of a heated aluminum plate in front of a large area blackbody as background. The viewing angle, defined as the angle between surface normal and camera line of sight, was varied by rotating the plate about its vertical axis. Stokes parameters were computed from the irradiance images. Images of intensity, degree of polarization and polarization angle were derived from the Stokes parameters. The dependence of these polarization characteristics on viewing angle was investigated. While intensity increased slightly with viewing angle, degree of polarization increased rapidly when the viewing angle was increased from 20o to 80o. The polarization angle increased with viewing angle and became constant at 150o for viewing angle greater than 60o. Target to background contrast based on degree of polarization increased with viewing angle and was higher than intensity contrast for viewing angle greater than 20o. Image processing algorithms were developed to segment the target plate from its background. The target similarity metric used was the texture-based Fisher distance, which enabled the fusion of one or more data type. The performance of the fusion schemes was compared via their Receiver Operating Characteristic (ROC) curves, which were plots of segmentation accuracy against false alarm rate. Binary image of the target was obtained by applying a Constant False Alarm Rate (CFAR) threshold. Fusion of intensity and polarization data produced better segmentation accuracy and lower false alarm rate than intensity-only data, for plate at viewing angle greater than 60o.