|Abstract||Automated cartographic mapping is a potentially important use of high resolution satellite imagery. In such an application, it is necessary to extract edge and line features and then identify those of
cartographic significance. This process is complicated because individual images contain a myriad of transient features, such as surface-cover texture. In addition, features of significance, such as road networks, may appear intermittent because they
traverse areas that exhibit comparable reflectance.|
On the positive side, those features of cartographic interest can be expected to exhibit a level of invariance when viewed in a temporal sequence of scenes. In this paper, it is proposed that
multi-temporal analysis be considered as an auxiliary tool. In such a scenario, comparison, of the edge and line feature content of coregistered image sets would be undertaken with a view to retaining only those features possessing a predefined level
of scene-to-scene consistency.
A modified version of the edge detection algorithm of Nevatia and Babu (1980) has been developed and implemented. This approach is of particular value since each edge is characterized by two attributes, namely,
magnitude and direction. In addition, spatially contiguous edge pixels can be logically linked together to form chains. Two multi-temporal editingmodes are proposed and evaluated, one based on per pixel comparisons between scenes and one based on the
degree of chain coincidence. Both perform well in suppressing transient feature when applied to a Thematic Mapper data set covering a portion of southwestern Ontario. However, chain editing has two distinct advantages: a robustness to image
registration errors; and an ability to overcome some of the intermittency problems in linear features, such as roads.