<title>Optical edge detection and segmentation of SLC SAR images with spatially correlated speckle</title>

Abstract

We have recently presented optimal criteria for edge detection and edge localization in Single Look Complex (SLC) Synthetic Aperture Radar (SAR) images. By working on complex data rather than intensity images, we can easily take the speckle autocorrelation into account, obtain more accurate estimates of local mean reflectivities, and thus achieve better edge detection and edge localization than with operators known from the literature. After a review of the theoretical aspects, we here propose solutions for the practical implementation. In SLC images the Maximum Likelihood (ML) estimator of reflectivity is the Spatial Whitening Filter (SWF), which is used in both tests. It necessitates precise knowledge of the speckle correlation. We describe how it can be determined and discuss the consequences of inaccuracies. Two-dimensional edge detection can be realized with multidirectional sliding windows. The watershed algorithm permits the extraction of closed, skeleton boundaries, and thresholding of the basin dynamics efficiently reduces the number false edges. The optimal estimator of the edge position is computationally intensive, so we examine suboptimal methods which require far less multiplications. The edge localization stage can be implemented with active contours or Gibbs random field techniques. Some segmentation results are shown.© (1998) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.