Higher-Order Spatial Statistics: A Strong Alternative In Image Processing

The goal of the this work is twofold, namely: 1) to discuss a method for identification of linear 2-D nonminimum phase system using higher order cumulants alone, and to derive the almost-sure convergence properties of sample estimates of higher-order spatial statistics. The measure of almost sure convergence is obtained for the sample estimates of third and fourth order moments and cumulants. 2) The experimental cumulants are used as texture features, they are later, incorporated into classification schemes. Some applications using synthetic textures and aerial images are presented to illustrate great descriptive power of higher-order cumulants.