Combination of speed profile of accreting/deleting texture and occluding contour geometry in determining relative depth

Abstract

Accretion/deletion of texture has traditionally been considered a decisive cue to ground status in determining relative depth. However, accretion/deletion can also arise from self-occlusion due to rotation in depth, and the accreting/deleting region can be perceived as figural (Froyen et al., 2013; Tanrikulu et al., 2016). Moreover, as the degree of convexity of a contour increases on one side, an accretion/deletion region is more likely to be perceived as a rotating figural object (Tanrikulu et al. 2014, 2015, VSS). In our previous studies, 3D rotation was perceived despite the constant speed of the texture. Here we manipulate the speed profile of the accreting/deleting texture, and examine its interaction with the shape of its bounding contours. In Experiment 1, a multi-region figure-ground ("rotating columns") display was used, consisting of alternating light and dark regions in which dots were moving horizontally in opposite directions. In Experiment 2, the stimuli were made more similar to traditional accretion/deletion displays, containing just two regions separated by a contour. In both experiments, we manipulated the speed profile of the dots in both sets of regions (either constant, cosine profile, or intermediate). We also manipulated the degree of piecewise convexity (part salience) of the contour(s). In both experiments, contour geometry was found to have a stronger influence on figural status than speed profile. Especially in multi-region displays, even a small convexity bias on one side largely overrode the effects of speed profile. The results underscore the importance of contour geometry—and its influence on figural status—in determining depth from motion, which most current models ignore. We present a Bayesian model, combining speed profile and contour geometry, that accounts for these findings without explicitly treating accretion/deletion as a cue to ground. Meeting abstract presented at VSS 2017