A multifractal approach for extracting relevant textural areas in satellite meteorological images

In the latest years, the use of computer vision tools for automatically analyzing the large amount of data acquired by remote sensing has grown in importance and number of different applications, ranging from basic research to industry. However, images displaying natural phenomena, and especially turbulence, develop following complicated patterns which are difficult to segment and to analyze with those tools. In this paper, we discuss on the use of new image processing methods to describe complicated flow and flow-like quantities, in applications such as meteorology. Using infrared satellite images as an example, we show that we are naturally led to gain insight in the physical and geometrical properties of the observed complex structures. We analyze different processing techniques (multiscale texture classification and multifractal decomposition and reconstruction) issued from the so-called multiscale methodology. The efficiency of multiscale methodology lies on its ability of reproducing known, experimental physical properties of the systems in study (such as scale invariance or multiscaling exponents) in the analysis scheme of images. We show that this methodology can be further exploited in order to derive information about a dynamical property from still infrared images. Namely, the main goal of our study is to detect and characterize textural areas at which typical convective movements take place. For that purpose, we compare the actual graylevel distribution in images, providing information about the temperature distribution, and a synthetic graylevel distribution induced by the multifractal formalism, also reinterpreted by its connection with thermodynamics. The conclusions of our work can be generalized to any analogous physical system.