Abstract
When four dot-pairs, rotating in phase about their respective pair centers, are arranged in a square configuration, they either group together into a "global" motion percept of two large squares translating along overlapping circular trajectories or appear as a "local" motion percept of four independently rotating dot-pairs (Anstis & Kim, 2011). The global percept is incompatible with, and perceptually very different from, the local percept, although the stimulus remains constant across perceptual switches. This type of motion-based perceptual grouping is interesting, because its output must serve as the basis for computing perceived motion, as global percepts are perceived to move more slowly than local percepts (Kohler, Caplovitz & Tse, 2009). Unlike Kanizsa-style inducers, dot-pairs contain no contours that share orientation with the illusory squares, which severely constrains the neural mechanisms that can be proposed to account for the grouping. Here, we exploited the fact that dot-pairs rotating out-of-phase are less likely to evoke a global motion percept, and presented subjects in the fMRI scanner with in-phase and out-of-phase dot-pairs. At each brief presentation, subjects reported whether their percept was local or global. We applied both univariate GLM analysis and multivariate pattern analysis within pre-defined visual regions-of-interest to find areas that distinguished between the local, rigid global (in-phase) and non-rigid global (out-of-phase) percepts. While we found no univariate effects, the multivariate analysis revealed that early visual (V1-V3) as well as motion sensitive areas (hMT+) carried information about the percepts. Classifying between physical differences in phase was not possible in most of these areas, indicating that image level features did not drive classification. We conclude that motion-based grouping of this type is mediated at least in part by mechanisms within regions of retinotopically organized visual cortex.
Meeting abstract presented at VSS 2012