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 imagery acquired from an airborne sensor flying over dense forest and sparsely covered terrain. We test for wide-sense stationarity, compute ensemble histograms, and estimate the autocovariance functions with associated error bars. We find that the statistics are stationary but non-Gaussian. Contrary to previous studies, we do not find that the correlations are described by decaying exponential functions. In fact, we find evidence for long-range correlations in the imagery, with autocovariance functions described by a relatively simple formula with power-law falloff.