Multiscale edge detection: a discussion on feature visibility

Centroid attraction is a multiscale approach to edge detection, which has been developed by the authors. It has been designed particularly for the analysis of synthetic aperture radar (SAR) data. We describe the method and explore one performance issue-that of feature size visibility. It is shown that in the simplest noise-free case, the feature must occupy at least half the filter window in order to appear in the filtered image. A number of further issues determining feature visibility in real data are presented. The visibility issue is illustrated using synthesized images, and centroid attraction is demonstrated by deriving an edge map from SAR data