Item

Abstract

When an object moves, it covers and uncovers texture in the background. This pattern of change is sufficient to define the object’s shape, velocity, relative depth, and degree of transparency, a process called Spatiotemporal Boundary Formation (SBF). We recently proposed a mathematical framework for SBF, where texture transformations are used to recover local edge segments, estimate the figure’s velocity and then reconstruct its shape. The model predicts that SBF should be sensitive to spatiotemporal noise, since the spurious transformations will lead to the recovery of incorrect edge orientations. Here we tested this prediction by adding a patch of dynamic noise (either directly over the figure or a fixed distance away from it). Shape recognition performance in humans decreased to chance levels when noise was placed over the figure but was not affected by noise far away. These results confirm the model’s prediction and also imply that SBF is a local process.