Abstract
Integrating visual features into objects correctly, known as the binding problem, is a fundamental challenge for the visual system. Here, we used the steady-state misbinding of color and motion (Wu et al., 2004) and fMRI MVPA to explore the neural mechanisms of active feature binding. The stimulus contained two sheets of random dots, one sheet moving up and the other moving down. On both sheets, dots in the right-end area and those in the rest area were rendered in different colors (red or green). When subjects fixated at the center of the stimulus, the color and motion of the dots in the right-end area were perceived to be bound in the same (i.e., misbinding state) or opposite (i.e., correct binding state) fashions as those in the rest area. During each fMRI run, subjects fixated at the center of the stimulus presented continuously for 180s and pressed one of two buttons to indicate their perceptual states, either the misbinding or the correct binding state. We trained a linear classifier to distinguish the multivariate patterns of BOLD response to these two perceptual states. Our results showed that the two states could be accurately decoded from the multivariate patterns as early as in V2. V4 and MT+ also showed performance above chance level. Furthermore, we trained another classifier with the multivariate patterns of responses to two physical (correct) binding stimuli (on each sheet, dots were rendered in the same color). We used the patterns responding to the misbinding states as the test data and found that only the patterns in V2 that encoded the physical binding predicted subjects' perceptual states, suggesting that the perceived misbinding stimuli were processed similarly to the physical correct binding stimuli in V2. These findings provide new evidence for the active binding of color and motion in human V2.
Meeting abstract presented at VSS 2016