Object-centred representations assign a reference frame to each object. Since an object model does not specify anything about the dependency of an object's appearance on viewpoint, recognition and location of an object requires invocation of the model (perhaps by some invariant characteristics) and determination of the relative position of object and sensor from the image or images at hand.

Viewer centred representations (VCRs) make the appearance of the object more explicit. Objects are represented by a number of views, each of which is associated with the relative position of object and sensor in space. In the most popular form of VCR , the aspect graph, the space around an object (the visibility space) is partitioned into regions, from each of which a set of visual features are visible. Crossing the boundary between two regions makes visual features appear or disappear from the image.

Exact algorithms have been reported to compute the aspect graphs of convex and concave polyhedral objects, and of solids of revolution. Although only a limited number of implementations have been reported which can compute exact visibility spaces for general objects, approximate visibility techniques have been used for several visual tasks, such as object recognition, robot planning and inspection.