Abstract
In classic conditions surfaces that undergo accretion or deletion of texture are always perceived to be behind a comparison surface. This is the case whether surface texture elements are moving or static, and whether the boundary between surfaces is moving or static. The region that moves together with the boundary becomes the foreground, and the background undergoes accretion or deletion. However, in the recent 'Moonwalk illusion' described by Kromrey, Bart & Hegdé (PLoS ONE 6(6), 2011) this is no longer true. Here the surface that undergoes accretion and deletion appears to be in front of the comparison surface. The change from classical conditions is the use of random flickering noise for the comparison surface instead of a coherent static or moving region. [demonstration at www.neu.edu/cvl] The ForMotionOcclusion model (Barnes & Mingolla, Neural Networks 37:141-164, 2013) includes two streams for computing motion signals and boundary signals. The motion stream first computes elementary motion signals; these are smoothed with a center-surround filter, which then supports the computation of accretion and deletion signals. The static stream computes boundaries based on contrast but also includes information about the locations of motion generated boundaries. The two streams generate depth percepts such that accretion/deletion signals together with boundary signals code for the 'far' depth. Plain motion signals code 'near' depth. The default code is for a 'middle' depth. The model fits the classical data as well as the observation that moving surfaces tend to appear closer in depth. Because the model separates the computation of boundaries from the computation of accretion/deletion signals, it also explains the counter-intuitive moonwalk stimulus. The lack of a distinct boundary due to the flickering makes accretion/deletion a non-effective cue for depth ordering. We show simulations demonstrating this result.
Meeting abstract presented at VSS 2014