Abstract
3D shape perception depends on multiple cues, which can be weighted according to their reliability in order to provide statistically optimal estimates of depth (Trommershauser et al. 2011). Typical psychophysical tests of optimal cue combination seek to isolate two cues, and vary their relative reliabilities. This allows us to measure their relative weightings (Hillis et al., 2004) and how these vary with changes in the reliability of each cue (Keefe et al., 2011). In typical natural scenes, many cues will be available, and the weighting of each cue will depend on both the availability and reliability of all other cues. The current study measured the weighting of binocular cues in naturalistic scenes in which multiple cues, such as perspective, texture and shading, also contributed to the perception of depth. Observers were presented with a single object which had been 3D scanned and rendered at two distances of 50 and 96cm. We manipulated the effective depth from binocular cues by varying the simulated interocular distance between 0 and 2 times that of each observer. A ‘pushpin’ gauge figure was presented monocularly at multiple points on the object, and its slant and tilt were adjusted so that it appeared to lie flat on the surface. The best-fitting surface mesh was calculated. Slant and tilt settings were highly correlated with the ground-truth values calculated from the object scan. At the near distance, the depth range specified by observers’ slant and tilt settings increased with the simulated IOD with a gain of 9%. At the far distance, there was no change in the perceived depth range with simulated IOD. These results show a modest but reliable contribution of binocular cues to the perception of surface shape for near objects. As predicted from geometrical considerations, this contribution reduced as the distance to the object increased.