The implementation of an object recognition necessarily requires the definition of a representation scheme for the objects. However, this choice not only determines the type of storage but even the manner of the recognition process - a relation which is similarly known in computer science as the connection between data structures and algorithms. Special attention shall be paid to the representation schemes because of considerable contradictions between the concepts of object centred descriptions and viewer centred descriptions.
Object centred models store a single representation for each object describing its properties. Accordingly, such a representation has to be three-dimensional. That is why many object centred descriptions resemble models used in computer graphics or computer aided design (CAD).
In contrast, viewer centred models contain collections of representations for the different perspectives under which the object appears to a (virtual) viewer. Thus a self-contained description of an object is embodied in the combination of its views. The stored views can be three-dimensional as well as two-dimensional.