An efficient two dimensional moving average model for texture analysis and synthesis

Abstract

A linear 2D moving average model which serves a dual purpose in the analysis and synthesis of stochastic texture is presented. This model is based on the assumption that a stochastic texture is produced by convolving an uncorrelated random field with a 2D filter which completely characterizes the texture. A simple and effective algorithm, involving the computation of arithmetic average, is developed to determine the parameters of the 2D filter from a given texture image. A major advantage of this algorithm is its insensitivity to additive noise, such as measurement error and quantization noise by which the image data is in general seriously contaminated. The 2D filter can be convolved with an uncorrelated random field to generate another image, distinctly different from the one given, that has the general visual characteristics as well as the first- and second-order statistics of the original image. Several synthetic images of real-life textures generated by this method are presented.<>