Evaluation Testbed for ATD/T Performance Prediction (ETAPP)

Currently, automatic target recognition (ATR) evaluation techniques use simple models, such as quick-look models, or detailed exhaustive simulation. Simple models cannot accurately quantify performance, while the detailed simulation requires enumerating each operating condition. A need exists for ATR performance prediction based on more accurate models. We develop a predictor based on image measures quantifying the intrinsic ATR difficulty on an image. We have implemented several image measures, including: CFAR, Power Spectrum Signature, and others. We propose a two-phase approach: a learning phase, where image measures are computed for a set of test images, and the ATR performance is measured; and a performance prediction phase. The learning phase produces a mapping, valid accross various ATR algorithms, which is even applicable when no image truth is available (e.g., when evaluating denied area imagery). The testbed has plug-in capability to allow rapid evaluation of new ATR algorithms. We present a performance predictor using a trained classifier ATR that was constructed using GENIE, a tool developed at Los Alamos. We have also formulated a measure that quantifies the fidelity of synthetically generated infrared imagery fidelity. This measure describes how well a set of real imagery is covered by synthetic imagery samples.