SAR image segmentation by morphological methods

The presence of speckle, which may be modeled as a strong multiplicative noise, makes the segmentation of synthetic aperture radar (SAR) images very difficult. The usual gradient operators yield poor results, but robust operators have been developed specifically for this kind of images. From the edge strength map ('gradient image') created by such an operator, closed skeleton boundaries running through local maxima must be extracted. This can be achieved with the watershed algorithm. However, to reduce the number of false edges, the algorithm must be made less sensitive to speckle. In this article, we compare two different approaches -- watershed thresholding and basin dynamics -- and propose a new algorithm for the computation of edge dynamics. The improvement brought by basin and edge dynamics is illustrated on ERS-1 images of an agricultural zone.