Object versus viewer
or
viewpoint independencies versus viewpoint dependencies

In the figure below, the pattern at the top is readily interpreted as a black object partly occluding a red object. Furthermore, this red object is readily taken to have shape A, even though shape B is possible just as well. Why is this?




Some approaches focus on the socalled T-junctions where the visible surface edges meet, and they take these junctions not only as cues that occlusion is at hand but also as cues for how the visible part of the occluded object is completed into a whole object. Such approaches, however, seem too simplistic. T-junctions are neither necessary nor sufficient for occlusion, and are merely cues for segmentation --- see T-junctions.

More sophisticated approaches take into account all candidate whole objects, to decide whether occlusion is at hand and, if so, which whole object is the best candidate. The selection of the best candidate may be guided by, for instance, the simplicity principle or the likelihood principle.

Before 1980, both paradigms focused primarily on viewpoint-independent aspects of objects. Then, to explain the example above, the simplicity paradigm would have to show that object A has a simpler shape than object B, and the likelihood paradigm would have to show that objects with shape A occur more frequently in the world than objects with shape B.

In the 1980s, the likelihood paradigm shifted to viewpoint-dependent aspects. For instance, in the example above, an object with shape B seems unlikely because, to yield the pattern at the top (which is what a viewer sees), it would have to take a very coincidental position with respect to the black object.

Since the 1990s, both paradigms combine viewpoint-independent and viewpoint-dependent aspects. This implies that object shapes and relative object positions are both quantified in terms of either complexities or probabilities, and that the combination determines which, either or not partly occluded, whole objects are predicted to be perceived.


For a further demo on these issues, see Occam, von Helmholtz, and Bayes
For a brief discussions of these issues, see In the Mind's Eye 2007 and Acta Psychologica 2011
For an extensive discussion of these issues, see Psychological Bulletin 2000

For a demo on the problem with quantifying probabilities, see Bertrand's paradox
For an occlusion model in terms of complexities, see Perception 1994