Texture analysis by universal multifractal features in a polarimetric SAR image

Multifractal features have been widely used in geophysics to characterize natural phenomena but seldom in image processing. Therefore, in this study, the authors show that these features can also be efficient in texture classification. They assume that a natural image texture is the result of a specific two-dimensional multifractal cascade process and so, according to the universal multifractal theory, few parameters are enough to entirely describe this process and therefore the texture. As the multifractal parameters are quite linearly separable, a very simple and fast algorithm is run for the supervised segmentation. In practice, the authors have computed these features on a SIR-C L-band polarimetric SAR image, and used them to segment this image into five classes (sea, shore, forest, urban areas and cultivated fields)