Estimation of noncausal stochastic model for a random image by means of `whiteness´

The authors propose a new, simple method for estimating a noncausal filter using the concept of the `whiteness´, and obtain the noncausal model parameters by minimizing the `whiteness´ of the output of a spatial filter. The method is successfully applied to several simulated images and practical textures, thus demonstrating its utility. The likelihood functional for a noncausal model, on the other hand, takes on a complicated form, and its maximization involves a troublesome computation when compared to the simple method using `whiteness´. Determination of model size by means of the `whiteness´ is also shown to be more effective than AIC or BIC in the case of image model estimation