Stochastic Image Modeling Using Cumulants With Application To Predictive Image Coding

Stochastic image representation techniques traditionally rely upon second-orc[er image statistics to determine the model parameters. This paper reports on the use of third-order cumulant statistics to imp] ement non-causal, phase sensitive ARMA image models. A novel, linear equation based 2-D MA parameter estimation algorithm is extended from an existing 1-D algorithm, for modeling colored prediction-error residuals, and is used in conjunction n with a non-causal 2-D ARMA parameter estimation algorithm. A weighted Ieast-squmes MA approach is also developed as has been done in the 1-D case. Application of cumtdant-based stochastic image representations to predictive image co,iing is discussed and preliminary results using causal and noncausal ARMA predictors are presented.