Application of multifractal analysis to segmentation of water bodies in optical satellite images

Abstract

In this paper, we study the characteristics of multifractal spectra obtained for remote sensing images through the coarse theory and propose a method for segmentation of water bodies based on it. In the first place, spectra of self-similar images created with Iterated Function Systems are calculated and compared with their statistical spectra. Then, optical remote sensing images are studied, emphasizing the differences between their spectra and those obtained for synthetic images. Attention is focused on the concavity of real image spectra and on its interpretation in terms of the analyzed images. This led us to the proposition of a segmentation method for water bodies. The method is tested and the results are compared with water masks of the regions under study. Comments are made about the limitations of the proposed method.