DECISION RULE FOR THE CHOICE OF GAUSSIAN OR LOGNORMAL MODELS FOR IMAGES.

The choice of whether to use autoregressive (AR) models to represent intensity or density data in image processing applications has often been made on heuristic grounds. The authors give a quantitative decision rule for selecting between AR models for intensity versus density. They illustrate the usefulness of the decision rule by testing Seasat synthetic aperture radar imagery, conventional photographs of textures, and real-world scenes. Comparison is made between the decision rule results and results of goodness-of-fit tests.