Multiresolution analysis scheme for ERS-1 image segmentation

We exploit the speckle information to provide ERS-1 images segmentation into regions, with reference to roughness. We describe a new segmentation scheme by using multi-scale information related to multifractal dimensions. We show that two different natural regions will give on radar imaging different roughness. The roughness characteristic may be considered as preliminary segmentation criterion. The natural measure of roughness is the fractal dimension. But the main characteristic of a fractal is its self -similarity. To keep the scale invariant pattern, a multiresolution technique is then incorporated to reduce the noisy roughness. This method associated to local fractal dimensions provides a good preliminary roughness segmentation scheme.