Texture simulation and analysis in coherent imagery

Coherent images or natural scenes, in particular those of the land using a synthetic aperture radar (SAR), often possess some kind of textural properties. It is desirable to be able to characterise such textural properties in terms of a model in order to provide an analytical means of detecting texture anomalies. The approach taken in the paper to understand these textural properties is a phenomenological one. Thus the use of a statistical noise model to describe clutter texture, which assumes that real data can be found that is homogenous, is proposed. The stationarity approximation allows real data sample estimates of statistical measures to be used to characterise the textures. Simulations using the model and parameter estimates are useful for assessing the worth of the model and how applicable it is. They also allow target detection algorithms to be assessed in a controlled manner. Usually only the single point and the first two point statistics are used to specify a texture. The importance of higher order two point statistics is investigated and measurements made using two particular texture simulators. How well these simulations match the higher order correlation properties of real data is considered