Texture Characterization And SAR Image Segmentation

Synthetic Aperture Radar (SAR) systems are powerful too;[& for analyzing the Earth's surface from a remote sensing platform. They present many interesting features, but provide images of the examined scenes that are affected by a high degree of noise, the socalled "speckle" effect. Usually, in the most cases, the first step performed when one has to process an SAR image is the filtering one, aimed at reducing the speckle effect. Many algorithms to this end have been proposed in the literature, but the feeling is that methods without disadvantages are not yet available in this specific applications. Nevertheless, the speckle effect is quite annoying, and some tools must be applied in order to enhance the information. In this paper, a new approach to SAR image segmentation is proposed, based on the assumption that, because no optimal filter is usable, it is preferable to analyze the original noisy data rather than information distorted in some way. Moreover, because an efficient way to characterize natural surfaces is to measure their textural features, from the original pictures textural images are extracted. This fact is aimed at providing the segmentation module with more data about the examined scene. In this way, the filtering step is by-passed by analyzing more and more information, so reducing notably the speckle dependence of the original data. To the end of extracting "virtual" images (obtained by the "physical" ones by means of some numerical transformatictns), the analysis has been focused on the analysis of the histogram shapes and of the multiple fractal dimensions. Several texture features are extracted from the local histogram: the signed deviation between the rank-order-filtered gray-value function and the equally distributed histogram; the average deviation from the locally representative gray value (i.e., the gray level of the more important peak); the histogram's degree of fragmentation (the latter parameter has proved to be very useful in the enhancement of thin details). With respect to the multifractals, they are a further improvement of the classical fractal geometry, a powerful tool for the characterization of natural surfaces. The main drawback of the single fractal dimensions is that they measure the surface roughness only, while other characteristics have to be evaluated, such as arrangement, spatial distribution This work was supported by the Italian Ministry of University and Scientific and Technological Research