Competition in multistable vision is attribute-specific

Ambiguous figures are visual stimuli with multiple competing interpretations. The different possible interpretations are mutually exclusive, and perception fluctuates from one interpretation to the next over time. Natural scenes are inherently ambiguous as well. A pair of two-dimensional retinal images does not instantaneously provide sufficient information to deduce the full three-dimensional layout and identities of objects in the scene. We investigate how the visual system uses internally stored models of objects and scenes to interpret and understand visual information. Ambiguous figures provide an experimental microdomain in which to investigate this top-down process in vision.

In one set of experiments, we explore the character of an apparent motion aftereffect, using a two-frame apparent motion stimulus that has two alternative interpretations as either left-right or up-down motion. This is termed the orthogonal axis aftereffect (OAE) and it is contrasted with the classical motion aftereffect (MAE). We find that the OAE, like the MAE, is increased with longer adaptation times. But, unlike the MAE, the OAE is position invariant, does not allow for multiple simultaneous aftereffects, and is attention dependent. These results are consistent with the hypothesis that the OAE involves the invocation and adaptation of a global scene model that is actively applied.

In another group of experiments, we explore the dynamics of perceptual competition when there are multiple ambiguous properties present in a stimulus. The stimuli are designed to have at least two different ambiguous properties in these experiments. We show that different ambiguous properties may alternate asynchronously, suggesting that competitive mechanisms are distributed and independent. In another experiment, the stimuli are designed so that the different ambiguous properties are linked, and so cannot switch independently. In this case, we show that one or the other properties will be chosen to stabilize perception, depending on the level of evidence for each property. Finally, we also show that a high-level ambiguous property---in this case, sense of rotation in depth---can serve to stabilize vision during binocular rivalry. Together, these experiments help to illuminate the nature of multistable perception, including binocular rivalry. Our experiments are evidence that perceptual competition is distributed and attribute-specific.