Supervised classification for synthetic aperture radar image

This paper deals with the supervised classification of synthetic aperture radar (SAR) images. Our approach is based on two criteria, which explicitly take into account the intensity of the SAR image and the neighborhood classes, similarly to the Pots model, but weighted by a discontinuity map. The high level of noise involves numerous classification errors, then we classify a restored image filtered with a well-adapted algorithm to clustering. Moreover, we isolate the texture of SAR images in order to help the classification. Finally, we present results on real SAR images.