ABSTRACT
Remotely-sensed imagery has for many years provided a means of classifying land cover over wide areas and at regular temporal intervals , and has, therefore, enabled the detec tion of broad changes in patterns through time, largely at the pixel level (Cope Bowley and Piper, 1 995) . Such imagery also has the potential for detecting more detailed in stances of change through identifying objects in a scene such as buildings and roads , thereby creating the possibility of updating cartographic databases. The problem is that, until very recently, commercially-available satellite imagery was l imited to a maximum spatial resolution of 1 0 m, which led to difficulties in extracting man-made features . This requirement for finer pixel resolutions when extracting components of land use was made clear by Blamire and Bamsley ( 1 996). The imminent an-ivaI of high resolution satellite imagery will allow more sophisticated, automated, object-extraction techniques to be implemented, where objects with characteristic shape, dimensions and interrelationships become the focus of attention rather than pixels .
