Mumford-Shah Filtering and Urban Area Classification

Abstract:
One of the very basic problems in remote sensing of urban areas is the delineation of urban areas themselves, useful for example in urban sprawl monitoring, or land use change inventory and planning. It is well known that classification of Synthetic Aperture Radar (SAR) images is made somehow difficult by the strong noise component due to speckle. Pre–filtering SAR data often results in a disruptive erasing of the embedded edge information, so critical for segmentation and classification. In this paper, we report on some experiments concerning the use of a variational approach, known as Mumford-Shah (M-S), as a smart filtering pre-processing block able to clean SAR images before (and compatibly with) their use in delineating urban areas through classification.