Entropy estimation and multiscale processing in meteorological satellite images

A new model for the multiscale characterization of turbulence and chaotic information in digital images is presented. The model is applied to infrared satellite images for the determination of specific areas inside the clouds. These images are difficult to manipulate however due to their intrinsically chaotic character, consequence of the extreme turbulent regime of the atmospheric flow. In this paper we briefly review some known techniques for processing such data and we will justify the necessity of multiscale methods to extract the relevant features. In the theory presented herein, one main attribute is determined for even, image: the Most Singular Manifold (MSM, of fractal nature), characterizing the sharpest changes in graylevel values. We will see that the most important set (from the statistical point of view) is that which both contains the sharpest transitions (MSM) and maximizes the local entropy. For that reason, images can be reconstructed to a good quality from the value of the gradient over that set of maximal information. The results are interpreted according to their relevance for determining meteorological features.