SPATIAL STATISTICS OF NATURAL-TERRAIN IMAGERY .1. NON-GAUSSIAN IR BACKGROUNDS AND LONG-RANGE CORRELATIONS

We report on an analysis of statistical correlations in midwave IR ima gery acquired from an airborne sensor flying over dense forest and spa rsely covered terrain. We test for wide-sense stationarity, compute en semble histograms, and estimate the autocovariance functions with asso ciated error bars. We find that the statistics are stationary but non- Gaussian. Contrary to previous studies, we do not find that the correl ations are described by decaying exponential functions. In fact, we fi nd evidence for long-range correlations in the imagery, with autocovar iance functions described by a relatively simple formula with power-la w falloff.