Adaptation of spatial filtering theory to the estimation of texture element size and regularity

A spatial filtering approach is adapted to provide a fast estimate of texture element size and spacing for ill-defined natural patterns subject to random or statistical variation. This information is useful for determining the optimum dimensions required for a spatial filter (or the receptive field of a neural network) when used for pattern discrimination. The method is demonstrated on samples from an international database and has potential use in the modelling, classification and segmentation of seabed textures and satellite imagery of forests and vegetation.