A transition from object centred to viewer centred models can be found in representations that make use of three-dimensional views. Biederman (1987) notices that his GSD could achieve only a limited invariance because certain feature arrangements are as a result of occlusion or self-occlusion not visible in all the projected views. Consequently he realized the necessity to store several views in the object representation. Thus his model is extended to hold a GSD for each characteristic view (similar to the concept of Freeman and Chakravarty (1980)), which is invariant within its individual range. A stronger emphasis on the recognition process can be found in the recognition by alignment introduced by Ullman (1989). It also deals with the representation of objects by three-dimensional views. However, recognition requires the determination of corresponding feature points between the input image and the stored models.
The main advantages of viewer centred representations originate from the fact that they can even be built upon two-dimensional instead of three-dimensional views as the following examples will show.