SAR Image Segmentation based on Multifractal Features

Abstract

Synthetic Aperture Radar (SAR) imaging is based on airborne or satellite active microwave sensors that can capture the earth surface by emitting a signal and receiving the backscattered signal that forms the resulting image.Since microwave radiation is not interfered by sunlight and can pass through clouds, SAR imagery can be generated oblivious to weather and daylight conditions.However, the active nature of the imaging process determines that SAR images are contaminated by an inherent {em speckle} noise that may degrade significantly the quality and usefulness of the images, and specific noise-removal processes may also filter out relevant textural information.In this article, we propose a texture-based method that can be applied for region segmentation in SAR imagery.The method is based on local analysis of the multifractal spectrum and a clustering procedure.The outcomes obtained both with synthetic and real SAR images show better region segmentation results than with state-of-the-art proposals.