How "intelligent" is the index tracking process?

We use the term "intelligent" here in the sense of Rock (1983) to refer to perceptually top-down or apparently global effects that constrain perception in such a way as to make vision more veridical in our sort of world (providing what Marr has called "Natural Constraints"). What we are interested in discovering is whether the indexing and tracking process is so low level that it precedes Gestalt-type patterning processes or the recovery of 3D properties. Although indexing is hypothesized to be a very early stage of vision, since it precedes the allocation of focal attention, this does not prejudge whether some of the very early visual computations -- such as those studied by Rock (1983), may precede it. Following the conservative strategy of other attention theories (e.g., Treisman's Feature Integration Theory and other selective theories from Broadbent on) we have not specified in any detail what the input to the indexing process is. Since it appears that at least the computation of 3D shape precedes processes occurring over a Feature Map (Enns & Rensink, 1991), it is not implausible that indexing and tracking also operates over a representation that has explicitly uncovered the 3D structure or which has already computed apparent motion or extrapolated occluded motion.