Region growing shadow segmentation in Synthetic Aperture Radar images

This effort introduces a noise-tolerant algorithm for segmenting shadow regions in Synthetic Aperture Radar (SAR) images, using both edge detection and a region-grow method for filling between detected edges. The technique introduced here recognizes that real shadow boundaries are neither precisely known, nor are infinitesimally thin, due to SAR phenomenology such as motion and layover. As the identified shadow region transitions from actual shadow to target and clutter regions, a metric based on the rate-of-change of the region size allows automatic detection of the shadow edge. This works in concert with traditional edge detection methods to identify boundaries more reliably. An evaluation of the quality of the shadow segmentation is presented using multiple quality measures, including a percent-pixels-same measurement that is based on manual segmentations of the same image. The algorithms and evaluation is accomplished on SAR target image chips obtained from the Moving and Stationary Target Acquisition and Recognition (MSTAR) program sponsored by the Defense Advanced Research Projects Agency (DARPA) and the Air Force Research Laboratory (AFRL).