Abstract
The small size of neurons receptive fields in early visual areas means that the brain must integrate local information in order to get a global percept. Often local information in the image gives rise to some ambiguity, allowing us to create illusory stimuli that offer unique ways to study the brain's “local-to-global” processing. One example is visual perception of the motion of an ellipse rotating around its center, first studied by Wallach (1956). Just by changing a single stimulus parameter (related to the aspect ratio) one can change the perception from “wobbly deformation” (“1D” motion) to rigid rotation (“2D”). We have studied how the perception varies between these two extremes, both psychophysically and computationally. For psychophysics, we created stimuli that enable us to estimate the amount of perceived rotation (R') by adding “beads” to the contour of the ellipse. For the beads, physical rotation R and perceived rotation R' are the same. When the physical rotation R for the added beads is similar to the perceived rotation R' for the ellipse alone, the beads have little effect on the observed motion. However, when R for the beads is different from the perceived R' for the ellipse alone, adding the beads induces a dramatic perceptual transition. We used this to measure the perceived rotation of ellipses of different aspect ratios. Hildreth's (1984) model predicts both rigid and wobbly perceptions by imposing a “smoothness” constraint: the motion of neighboring points is required to vary as little as possible from one point to the other, while keeping it consistent with the local information received. Using this theory, we predicted the perceived rotation R' for physical rotations of ellipses with different aspect ratios, and compared these predictions with the psychophysics. The results were similar, though there was one significant difference: the model tends to predict, for intermediate ellipses, a lower perceived rotation than what found experimentally.
PJV gets support from the Swartz Foundation