SAR segmentation algorithms: a quantitative assessment

This paper compares the performance of two segmentation algorithms, the MUM (Merge Using Moments) and RWSEG, using simulated synthetic aperture radar (SAR) data. This comparison is performed from both the qualitative and quantitative points of view, and carried out using synthetic images that contain regions with different homogeneity degrees, aiming at land use applications. The process for obtaining simulated images consists of creating an idealized classes image (a phantom), which summarizes the main geometric and topologic characteristics of Amazonian targets. Then a statistical model of observations from each class is proposed, and a Monte Carlo experience is performed.