Abstract
Visual motion is a critical behavioral cue. Contemporary neural models estimate motion by computing pairwise space-time correlations in light intensity. Moving natural scenes, however, contain more complex correlational structures. By simulating motion using natural scenes, we show that specific third-order correlations resulting from asymmetries in above- and below-mean regions of the visual scene contain useful information about motion. Moreover, motion estimation models that utilize odd-ordered correlations are able to distinguish between light and dark edges, something that 2nd order models cannot. Given that this information exists in moving natural images, we asked whether third-order correlations are estimated in a manner that distinguishes moving light edges from moving dark edges. First, to isolate light- and dark-edge specific neural responses, we used novel stimuli that separately manipulated motion direction and edge polarity. Using these stimuli and Steady-State Visual Evoked Potentials, we demonstrated that humans exhibit adaption that is specific to the combination of edge direction and edge polarity. Second, to isolate perceptual sensitivity to high-order correlations, we used 3-point "glider" stimuli that contain no net 2-point correlations. These stimuli separate the motion information contained in 3rd and higher-order correlations from that specified by 2nd-order correlations and they produce a percept of motion. To test for the connection between these high-order correlations and edge-polarity specificity, we first adapted participants to moving light and dark edges and then measured psychophysical sensitivity to the 3-point "gliders". We found that this adaptation modulates the perception of 3-point gliders. Our results thus indicate that a computation of high-order correlations underlies edge-polarity selective motion processing.
Meeting abstract presented at VSS 2013