Abstract
A picture creates the retinal image of the original 3D scene when it is viewed from its center of projection (CoP). When the picture is viewed from an oblique position, the retinal image is distorted. Yet the perceived spatial layout seems similar to the depicted layout. Pictorial-compensation hypothesis (e.g., Kubovy, 1986; Sedgwick, 1980) suggests that geometric cues in the picture (e.g., rectilinear structure in the depicted scene) are used to recover the original CoP. This is then used to compensate for oblique viewing, generating a percept consistent with the original scene. Alternatively, compensation for obliqueness might result from transformation of the retinal image by the estimated slant of the picture's surface, in a process akin to shape constancy; we call this the surface strategy. We tested the pictorial-compensation and surface-strategy hypotheses by having observers judge sphericity and surface slant in pictures. The key manipulation was varying projection angle (angle between the projection surface and the projection axis) and viewing obliqueness independently. We varied available cues to pictorial surface slant, as well as geometric cues in the scene that could be used to recover the CoP. Stimuli were ray-traced images of a scene containing ellipsoids and/or rectangular surfaces. For the ellipsoid, observers adjusted aspect ratio until it appeared spherical. For the rectangular surface observers adjusted slant until it matched a criterion. The amount of available surface information had a large effect on observers' settings. Under binocular viewing, neither available geometric information nor projection angle had an effect, which is consistent with surface-strategy and not pictorial-compensation hypothesis. Overall, the results do not support a scene-wide pictorial-compensation hypothesis. Instead, effects of oblique viewing may be due to transformations of retinal image by the slant estimated locally across the pictorial surface.
supported by NIH NRSA (DV) and DOE CSGF (ARG)