Templates with lower mean square error for ATR

Many automatic target recognition systems work by matching test profiles against profile templates from various parts of the targets. These templates are generally constructed for uniform-width sub-targets constructed out of the targets being considered. In this paper we present a heuristic search based algorithm for constructing optimal-sized sub- targets which may be of varying widths. We show the improvements that ar obtained by constructing templates for sub-targets identified by the search algorithm. The reduction in mean square errors is reported and also the performance of two classification test runs.