Texture modelisation by multifractal processes for SAR image segmentation

The introduction of the fractal geometry a few years ago brought up a new approach in texture analysis and classification: a surface can be described by its roughness which is quantitatively evaluated with its fractal dimension. This feature has been widely used, with quite good performance, but some authors have pointed out that the single fractal dimension may not be a successful tool for fully discriminating textures; some studies on different textures showed that despite obvious visual differences, the fractal dimensions remained quite identical. So, we attempted to find a more accurate description of the texture. In remote sensing, the computation of the multifractal parameters of some physical phenomenons like clouds or sea ice, pointed out a certain stability of their value whatever the remotely sensed data sources are. A study on multifractal parameters of interest in classification is performed on the SIR-C L-band polarimetric SAR image of the Landes area in France. We introduce the multifractal parameters and then show how it can be used for texture discrimination. Finally, parameters pictures, and classification results obtained using the multifractal features are presented.