Effective use of the large amounts of timely data to be provided by the coming generation of space-based radars will require automatic methods of image interpretation. The key to such interpretation is an image representation, based on low-level operations, which can support the introduction of high-level (rule-based) knowledge. This representation, described in this paper as a segmented image database (SID), is dependent on the performance of the low-level operations (segmentation, edge-detection and thin-line-detection) which generate it. Methods of quantifying the performance of these operations are described. Use of the SID to support classification based on context, and image-map matching that uses image structure, rather than geometrical matching, are demonstrated.