Abstract
It is an open question how different cortical areas interact to accomplish the robust analysis of moving visual patterns. Here, we present a model of recurrent interaction of V1, MT and MSTd utilizing a space-variant mapping of the retino-cortical pathway. We claim that a recently discovered motion illusion (Pinna & Brelstaff, Vis.Res. 40:2091–2096, 2000) unravels the underlying mechanisms involved in processes of motion integration and segregation.
The (static) pattern that leads to a strong illusory motion effect consist of two rings of diamond-shaped items each outlined by light and dark lines such that coarse contrasts appear along the diagonal axes of individual items. While fixating the center of the rings an approaching viewer perceives the rings as rotating in two opposite directions. An essential part of our model is the transformation of the retinal input flow pattern to a log-polar V1 representation (Schwartz. Biol. Cybernetics 37:63–76, 1980) achieving a mapping of the peripheral true motion into a horizontally oriented uniform gauge pattern. Motion information from two successive frames of a movement simulation are integrated along the V1-MT-MSTd feedforward pathway utilizing direction selective cells of increasing spatial size (X:Y:Z ratio; Grossberg et al. Cerebral Cortex 9:878–895, 1999).
The illusion is shown to induce a motion pattern in the cortical MT and MST areas. MST-MT feedback achieves the necessary disambiguation of initial noisy estimates. This leads to segregated opponent motions along circular directions when perceptual splitting occurs, while homogeneous motion fields occur when no perceptual splitting is observed. The results provide evidence for the essential role of feedback from higher-order motion area MST.