Preparing New Internet Reading Courseware: READY 2000

Not Available We propose a comprehensive program to develop a mathematical characterization of clutter environments in SAR imagery. In contrast to conventional characterizations of SAR clutter, which use simple statistical models for targets and background to predict the receiver operation characteristics of front-end detectors (e.g., CFAR algorithms), we develop a more general notion of clutter complexity that also considers the significant screening power afforded by target type-specific back-end processing. Our complexity measure may be computed over sample clutter image sets, and used to predict the false alarm rates likely to be incurred by generic ATR algorithms over comparable imagery. Our approach is based on constructing estimators for the density of target-like objects in the clutter set. In Phase I, we develop a range of estimators that provide different computational cost vs. estimation performance trade-offs, and empirically evaluate their performance. In Phase II we will extend our methodology to better address situations where inadequate example clutter imagery is available on which to base good empirical estimates of complexity. We expect that our research will provide useful tools to support ATR evaluation and performance benchmarking, as well as insights and theoretical advances that