Given the primitive edge elements extracted by a localised edge detector, it is often necessary to group these edge elements into extended boundaries which may be used as part of a scene description or compared to wire frame models in a knowledge base. This is often referred to as boundary segmentation. If little or no prior knowledge of context is known, then the mechanism has to proceed on the basis of generalised rules, based on the local or global structure of the edge data. In this section, we consider two of the most widely applied techniques for grouping edges to form boundaries, the first of which is based on a local tracking strategy and the second on a global grouping based on common parameters. We then consider a third approach which falls between the two extremes. Having grouped edge pixels into extended boundaries, we then discuss how these may be represented in a a higher form for subsequent processing.