Principal component background suppression

We have developed an adaptable background suppression algorithm, based on the statistical technique of principal components, to mitigate the effects of sensor line of sight motion (clutter) across structured background scenes. The central idea is construction of a “background space” as a linear vector subspace modeling the background being viewed. We have applied our algorithm to two test cases which were constructed by simulating random motion of a staring array focal plane over a high resolution scene. The first test case, with clutter noise only, found a low-intensity signal (S/N=0.05) with a 245-fold enhancement by projecting out a background space using 40 principal components. The second test case added Gaussian electronic noise and found the signal with a 34-fold increase in signal-to-noise using 16 principal components. This is believed to closely represent the actual problem encountered in staring array focal planes. Our results show that increasing the number of principal components increases the algorithm´s ability to suppress clutter up to the point where electronic noise becomes dominant. We give a heuristic argument for determining the proper number of principal components for maximum signal-to-noise enhancement