Non-Gaussian stochastic model for analysis of automatic detection/recognition

Abstract

We demonstrate that a very simple stochastic model based on nonlinear transformation of Gaussian random fields can be successfully used to model homogeneous non-Gaussian natural backgrounds observed in a wide range of airborne and spaceborne sensor imagery. We use this model to simulate backgrounds ranging from IR forest terrain to SAR woodland and SAR sea surface imagery. The model reproduces the histogram, second-order correlations, and third-order correlations measured in the real imagery. We discuss applications in the design and analysis of algorithms for automatic detection and recognition of objects embedded in natural imagery.