Clutter Modeling and Performance Analysis in Automatic Target Recognition 1

The past decade has witnessed rapid development in accurate modeling of 3D targets and multiple sensor fusion in automatic target recognition (ATR), however, the scientiic study for quantifying non-target objects in a cluttered scene has made very limited progress, due to its enormous diiculties. In this paper, we study two important themes in ATR: I) clutter modeling { how can we build generic and low-dimensional probabilistic models for cluttered scenes, and how can we automatically learn such models from observed images? II) performance analysis { how can we quantify the eeects of clutter on the performance of the ATR algorithms, and how much do the learned clutter models improve ATR performance? We answer the above questions by combing two important trends which have emerged in the past few years. The rst is the minimax entropy learning theory, proposed by Zhu, Wu and Mumfordd12]. According to this theory, cluttered scenes are deened on random elds with features characterized by generic lters at various frequencies and orientations. Then a probability model is learned so that it reproduces the statistics of the most important features. The second is the theory for performance bounds (such as the Hilbert-Schmidt bound), proposed by Grenander, Miller and Srivastavaa3]. This theory quantiies the accuracy of target estimation on Lie groups. We illustrate our theories by three groups of experiments: 1) clutter modeling, synthesis, removal; 2) eeects of clutter on the performance bounds; 3) using a learned clutter model for ATR in contrast to an additive Gaussian noise model. 1 The authors are members of the Center for Image Sciences supported by a grant DAAH-04-95-1-0494 from the ARO.