Abstract
In pictures, outwardly directed objects (e.g., the pointing arm in the Army recruitment poster), seem to “follow” the viewer. Fronto-parallel objects (the shoulders in the poster), do not seem to follow the viewer. If objects slanted at ∼90 deg in the picture seem to follow the viewer and those at ∼0 deg do not, the perceived slant between pictured objects should change with viewing location, suggesting a failure of perceptual invariance. To investigate this, we used pictures of vertical hinges in an “open book” configuration. Observers viewed them from various azimuths and indicated the perceived angle between the sides relative to 90 deg. We varied the information for estimating picture-surface slant and geometric information in the picture background. We considered three hypotheses. 1) Retinal: Failure to take viewing location into account. 2) Perceptual-invariance: Perceiving the same 3d structure from all locations. 3) Weighted-average: Combination of slant estimates from the depicted scene and picture surface (this hypothesis is only partially consistent with invariance). With minimal picture-surface slant information (monocular viewing through aperture), the results were similar to the retinal hypothesis (no invariance). With rich surface slant information (binocular viewing, picture frame cue), they were similar to the weighted-average hypothesis (partial invariance). Varying the geometric information had no effect. The perceived layout of pictured scenes changes with viewing position, but much less so when information about the picture-surface slant is available.