Stochastic simulation of infrared non-Gaussian natural terrain imagery

Abstract

We describe a new technique for the fast simulation of natural IR imagery with non-Gaussian statistics. We pass samples of the imagery through an invertible pointwise nonlinear transform, to give samples of a new random field with Gaussian marginal probability density. We compute the correlation function for the transformed samples, and simulate Gaussian random fields (GRFs) with these correlations. Passing the simulated GRFs through the inverse transform gives our simulations of the original imagery. We discuss the effectiveness of this approach for simulating natural terrain backgrounds.