Combining watershed and statistical analysis for SAR image segmentation

Synthetic aperture radar (SAR) images are affected by speckle noise, which gives a granular appearance to them. This noisy pattern can lead to difficulties when dealing with tasks such as edge detection, image segmentation, classification and interpretation. This paper presents a new technique for SAR image segmentation that combines mathematical morphology and speckle noise statistics. The novelty of this approach consists on a modified version of the watershed transformation that discards a preprocessing step or data conditioning. Segmentation results on real SAR and synthetic speckled images demonstrate the effectiveness of the proposed method.