Abstract
Perceived 3D slant is often biased, especially when depth information is poor. A possible Bayesian explanation is that visual cues are integrated with a prior that is weighted toward a default (frontal), resulting in biases toward the default when cues are unreliable. By this explanation, slant perception should become less veridical as the cue reliability decreases. In a previous study, we observed that biases in perception of slant from texture varied with slant as expected based on cue reliability. In this study, we varied the reliability of stereo slant information by changing viewing distance (90 vs 180 cm), and measured perceptual biases in perception of slant from stereo and combined stereo and texture. Subjects were presented with images of slanted planar surfaces with surface texture that was either informative (Voronoi) or uninformative (broadband noise), and estimated the perceived surface slant by aligning the palm of their hand. Subjects also performed 2AFC slant discrimination to measure reliability of slant information. A Bayesian model predicts that when stereo information becomes less reliable at longer viewing distance, the amount of perceptual bias would increase and texture information would have more relative influence. In monocular conditions, we found that slant estimates were a nonlinear function of simulated slant, as observed previously, consistent with poorer slant discrimination at low slants. In binocular conditions, discrimination thresholds were larger with longer viewing distance, and slant estimates also showed more bias toward frontal. At near viewing distance, slant estimates varied linearity with slant, and were similar with Voronoi and noise textures. At longer viewing distance, slant estimates were less biased with the informative texture, and showed nonlinearity in the same direction as monocular texture conditions. All of these results are qualitatively consistent with a Bayesian model that integrates slant cues and a prior according to their relative reliability.