Viewer centred representations not only give rise to these successful technical implementations but even have a close relationship to biological findings. The influence of different vantage points on object recognition is a fact well known from psychological experiments in which classes of favourable and of disadvantageous view points were encountered. Yet, the existence of these so-called canonical and accidental views has diverging interpretations. Supporters of object centred representations might argue that accidental views coincide with mathematical singularities which impair the reconstruction of a three-dimensional model. However, this hypothesis cannot account for the continuously decreasing recognition rates reported by Edelman and Weinshall (1991), which show obvious dependencies from the distance to previously trained views. These results were obtained for monocular as well as for stereoscopic presentation.
The outcome of the psychological experiments may be explained by Perrett et al. (1991) on a neurophysical level. Single cell recordings in the macaque superior temporal sulcus (STS) showed some cells with clear viewer centred behaviour. They were only stimulated by certain perspectives of familiar faces. Another kind of cells reacted nearly uniformly over all the views of a person and can therefore be considered as an object centred representation. The discovery of two cell types does not unveil a contradiction but rather demonstrates that object centred outputs can be interpreted as the simple supposition of several viewer centred cells. Besides, viewer centred representations make also sense under behavioural aspects, as the relation of an object/subject to the viewer is of eminent importance for his or her current interactions and future deeds. Faces, for example, offer such a special information by the position of the eyes which gives a clue to the focus of attention of the opposite subject. In fact Perrett et al. report the finding of cells recognizing faces with sensitivity to different positions of the eyes. Nevertheless Perrett et al. (1987) emphasize that the principle of viewer centred representations verified for face recognition should not be regarded as a special case but as the foundation of a general object recognition scheme.
Tanaka (1996) states some further neurophysiological studies leading to similar conclusions. Moreover, he gives an account of face sensitive cells in another brain area, the anterior inferotemporal cortex (AIT), which not only react in a viewer centred fashion but even show a systematic arrangement with locations for neighbouring views being structured in columns.
Additionally, the neurophysiological evidence seems to reveal a universal principle: The search for a highly invariant recognition mechanism is solved in a way which represents a trade-off between memory and computation.