Non-uniform Correction Algorithm Suite (NUCAS)

Image noise is often a limiting factor in target detection and recognition using infrared cameras. Many techniques exist for limiting temporal noise, but the predominant solution for fixed pattern noise correction is radiometric non-uniformity correction (NUC). This adds moving parts to the camera design and requires taking the camera off-line every 10 minutes or so during operation. Our proposed scene-based (SB) NUC alternative uses real-time image processing to perform NUC continuously during normal camera operation. In Phase I we prototyped the best known SBNUC algorithms and evaluated them on video sequences exhibiting widely varying fixed pattern and temporal noise. We also characterized their performance as a function of image motion (due to camera or scene motion). Consequently, we identified the optimal operating conditions for three types of SBNUC algorithms: constant statistics, motion compensated temporal filtering, and spatial filtering. In Phase II we propose to develop image analysis algorithms capable of determining the camera’s current operating condition in real-time, and hence specifying the most effective SBNUC to use for each video frame. Through our partner Nova Sensors, our algorithm suite will be implemented as an FPGA-accelerated dedicated image processor for NVESD’s SE-IR cameras, enabling extensive system evaluations outside laboratory environments.