ABSTRACT
Extraction of information from remotely-sensed data is often carried out without any knowledge of structure in the imagery; pixel-based classification is an obvious example. For many purposes, however, it is necessary to recognize structures, such as agricultural fields or ice floes before the pertinent questions (has it changed? has it moved?) can be formulated. Segmentation is the process of splitting an image into 'homogeneous' regions and hence imposing a geometrical and topological structure on the pixels within it. The human vision system is adept at this operation. Indeed, we cannot prevent ourselves from segmenting the visual input. It appears to be an essential part of image understanding, possibly since it reduces the postretinal data rate to an extent that is manageable by the successive visual processing (Resnikoff, 1989). Survival requires organization and selective omission of data.
