Depth cue anisotropies and weights

Purpose. Structure-from-motion is more sensitive for a cylinder rotating about an axis orthogonal to the cylinder axis than about a parallel axis (Cornilleau-Pérès & Droulez, Perc. & Psychophys., 46, 1989). Sensitivity to stereo curvature is higher for a vertically- than for a horizontally-oriented cylinder (Rogers & Graham, Science, 221, 1983). Do either of these differences in cue reliability affect the cue weight in depth cue combination (Landy et al., Vis. Res., 35, 1995)? Methods. Stimuli were rotating, textured, elliptical half-cylinders viewed stereoscopically. The depth portrayed by each cue (texture, motion and stereo) was manipulated independently by using texture elements stretched along the viewing direction and by using a fallacious value of the inter-pupillary distance to calculate the ray-traced images. Inconsistent cue sequences (where one cue indicated a depth different from the other two cues) were generated for vertically- and horizontally-oriented cylinders, rotated about vertical and horizontal axes. Consistent cues stimuli were generated for horizontally-oriented cylinders rotated about a vertical axis. In each trial, subjects viewed one inconsistent and one consistent cues stimulus and indicated which appeared to have greater depth (temporal 2I-2AFC). The consistent cues stimuli are used as a "ruler" to measure perceived depth. Results. The increase in perceived depth per unit increase in depth of a single perturbed cue (in the inconsistent cues stimulus) provides the weight on the perturbed cue (Landy et al., 1995). We find no consistent pattern of cue weights as the cylinder orientation and axis of rotation are varied. This is not what one would predict from the curvature anisotropies. Individual differences in the pattern of results are significant. Conclusions. Observers do not consistently use cue reliability determined by stimulus geometry as a factor in choosing cue combination weights.