The perceived slant difference between twisted patches is known to be reduced by partial occlusion, a phenomenon known as slant assimilation (Fantoni, Gerbino & Kellman, 2004; Liu & Shor, 2005). Fantoni, Gerbino and Kellman (2008) demonstrated that this phenomenon is diagnostic of visual approximation, a process mediating Amodal Completion (AC) of 3D structures that causes a distorted representation of image-specified parts. Approximation-based distortions were previously studied in impoverished viewing conditions (immobile observer) and with impoverished stimuli (stereoscopic patches under orthographic projections). Here, we show that similar distortions occur in more general viewing and stimulus conditions, as when the observer is naturally moving and the twisted surfaces are specified by self-generated optic flows under perspective projections. The sensitivity to discriminate between rigid and non-rigid 3D structures was impaired when the twisted patches appeared as the unoccluded parts of a smooth surface partially hidden by a foreground frontoparallel surface, relative to cases in which they were perceived as separated patches, either in the absence of the foreground occluding surface (Experiment 1) or with the occluding surface in the background (Experiment 2). The exact same biases were also found for passive observers, who experienced from a static vantage point the same optic flows generated during active viewing. Results are compatible with a Bayesian model that disregards egomotion information and selects a reliable prior for stationarity/rigidity in presence but not in absence of amodal completion. Results are incompatible with a model predicting that image encoding is biased towards slant assimilation in presence of AC. This calls for an update of the current notion of visual approximation in AC.
Meeting abstract presented at VSS 2014