Edge validation in SAR images using likelihood ratio tests

SAR image segmentation is performed in four stages. The first three are conventional and familiar to many researchers: 1) reduce the multiplicative speckle noise; 2) detect edges between different surfaces; 3) clean and link edges to form regions. The fourth task is the authors' newly introduced statistical testing procedure for validating suspect edges, in particular, those linking edge segments. This step is necessary because the result of the first three will inevitably contain errors due to the inadequacies of noise filters and edge detectors, as well as the heuristic nature of the linking process.