Improving the update procedure for a digital topographic map data base in an operational environment is a non-trivial task. Typical update procedures involve manual change detection on (bi)temporal sets of aerial photos and/or direct comparison of the latest map with a newer photo. Full automation of this process is not within immediate reach. In the short term, however, much can be done to reduce the burden of manual updating, for example by introducing automated change detection methods. This

more »

... involves the comparison of the topographic data base (which can be thought of as a highly refined/generalized image) and a new, image, which must be segmented/generalized prior to the comparison. In an operational setup, this also involves long term storage of large anounts of image data, making lossy compression necessary. This compression must be used with great care in order to avoid negative influence on the spectrally based segmentation. Due to the complex nature of the mapping from topographic classes to spectral signatures, the segmentation must typically be carried out using some level of supervised classification. This is not necessarily desirable for operational tasks, so in this initial study, unsupervised clustering of a spectrally decorrelated colour photo into a very small number of classes is used, in an attempt of facilitating the transition to operational levels. This approach results in a quite meaningful set of clusters, but obviously not in something that can be used in a fine grained classification.