Automatic detection of large object features from SAR data

The feature extraction operation, from SAR images, encounters many difficulties due to the high level of noise and the poor quality of the object contour. This paper aims towards an implementation of an automated algorithm for extracting features of large objects such as rivers, lakes, or highways from SAR imagery. The method uses a watershed algorithm to detect homogeneous areas. The above areas are then grouped into probable objects using heuristic reasoning about the similarity between neighbouring regions and the difference between larger background areas. Results from SAR images of the Ottawa area are given