MULTIRESOLUTION STATISTICAL-ANALYSIS OF COMPUTER-GENERATED FRACTAL IMAGERY

A statistical methodology for the analysis of strongly structured stochastic fields is described and illustrated by application to computer-generated images derived from Julia, and related Mandelbrot, sets. Convolution filters, designed to detect elementary image structures such as bars and blobs, are applied over a range of scales. The results are displayed as statistical distributions, normalized to scale-invariant form using fractal dimensions which characterize the image. Further image characterization is achieved by the use of an order parameter which measures the degree to which local patterns in the data match the structures defined by the associated filters. Reference is made to analogous results obtained from an analysis of remotely sensed images of natural Earth backgrounds. © 1991.