Effects of misalignment on pixel averaging when scaling templates and images before scanning

Abstract

Discusses the effect of pixel misalignment when dealing with templates and scenes which have been scaled down in size. One situation where this may occur is when templates or training vectors arise from images acquired at one range, and the image under question has been acquired at a multiple of this range. Here one could scale the template accordingly before beginning to scan the scene. Another application is where the template and scene are at the same range, but their dimensions are so great as to make the scanning process overly burdensome computationally. Here one could scale both the template and scene by the same factor before scanning to reduce the number of computations. The results of this coarse-resolution scan would then give starting points and neighborhoods for finer-resolution scans. For either of these applications, if there is pixel misalignment of the scene compared to the template, then an error is introduced and the measure of closeness of match is degraded. This is caused by the fact that the boundaries over which the pixel averaging is performed may be different for the scene than for the template. The impact of such misalignment is investigated here. An analytical expression for this effect is obtained and a means is presented for evaluating it based on the template alone, without knowledge of the scene to be scanned. Simulations are performed to test the methodology.