Abstract
Recovering meaningful objects from the elementary retinal samples of a visual scene involves a cascade of binding processes unfolding along a visual hierarchy that splits into a dorsal and a ventral stream, each functionally specialized for the analysis of motion and form. Unified perception of moving forms thus requires interactions between these streams, but the nature and function of these interactions remain elusive. We address this issue using fMRI together with 3 versions of bistable moving ‘aperture’ stimuli whose interpretation fluctuates between a whole moving shape (bound state) and the independent motion of its parts (unbound state). We identified a dynamic balance of BOLD activity between dorsal and ventral visual areas tightly correlated with the dynamics of perceptual alternations: A transition towards a bound state is followed by a BOLD activation in V1 & LOC and a deactivation in hMT+. In contrast, a transition towards an unbound state entails a BOLD activation in hMT+, and a reduced activity in the LOC. During bound states, this network extends to V1, V2 & V3 suggesting the progressive involvement of recurrent LOC/V1/V2/V3 loops. This extended network is not seen during unbound states. Importantly, spontaneous alternations with unchanging stimuli and perceptual alternations induced by periodic changes of luminance, motion noise or shape, known to control form/motion binding, entail a similar network dynamics. Analyses of eye movements indicate that saccades and blinks are more frequent before transitions although overall too rare to account for BOLD contrasts. Altogether, these results suggest that ventral areas provide constraints to select motion components to be bound into a single motion within dorsal areas.
Supported by French Ministry of Research (ACI) “Cognitive Neuroscience” to JL.