Multiresolution hidden Markov trees for analysis of automatic target recognition algorithms

We present a method for characterizing clutter in images by using multiscale hidden Markov models in the wavelet domain, and we discuss the application of this in predicting the performance of automatic target recognition (ATR) systems. Wavelet Markov models explicitly include dependence across scale, which allows us to characterize highly structured clutter. Since structured clutter is a common cause of false alarms in modern ATR systems, these models give us the ability to analyze and predict false alarm rates and other ATR performance characteristics. These predictions can lead to adaptive ATR systems which automatically adjust to local clutter conditions.