Image modeling and quality criteria

When an image which is disturbed by additive noise is recursively filtered, the filter is derived from a dynamic model based on the second-order statistics of the original image and noise. For given image statistics, it is possible to consider different models. The choice of a model may influence the quality of the filter results. To compare different models representing the same image, three quality criteria, including the Cramér-Rao bound, are introduced. Results of this comparison are discussed.