Integration of stereo and texture slant cues measured with a slant estimation task

Abstract

Multiple sources of information are typically available for perception of 3D slant, such as stereo and texture. The relative contributions of different cues have been measured using a cue conflict paradigm. Some studies have found that stereo and texture contribute to perceived slant in a statistically optimal manner (e.g., Knill & Saunders, 2003; Hillis et al 2004). Todd, Christensen, and Guckes (2010) recently criticized the discrimination task used in previous studies. Todd et al used an estimation task to assess the perceived slant of cue conflict stimuli and observed no significant contribution from texture when stereo information was available. The goal of the present study was to investigate the perceptual weighting of stereo and texture cues using a slant estimation task. Stimuli were binocular views of a textured surface slanted in depth around a horizontal axis. The slant information provided by texture was manipulated by varying slant (0°-60°) and texture type (noise vs Voronoi). Observers indicated perceived slant by adjusting a 2D probe figure. Slant estimates showed large deviations from veridical, which varied across individuals, but tended to be systematic functions of actual slant. A subset of trials had conflicting cues: the slant specified by texture differed from the slant specified by stereo by ±15°. The relative weighting of texture and stereo information was inferred by comparing slant estimates in cue conflict and cue consistent conditions. We found that the Voronoi texture significantly influenced slant estimates, and the weighting of texture increased with slant. Our results are qualitatively consistent with those of previous studies using a discrimination task, suggesting that evidence for optimal integration of slant cues is not an artifact of methodology.