Effects of viewing geometry on combination of disparity and texture gradient information

The visual system estimates 3D scene layout from several information sources (or depth cues) in the 2D retinal projections. How is the information from different depth cues combined to form a single 3D estimate? We examined how binocular disparity and texture gradients are combined to estimate surface slant. Several authors have argued that depth cue combination may be modeled as a weighted average: Sc = wt*St + wd*Sd (St = texture slant estimate, Sd = disparity slant estimate, wt = texture weight, wd = disparity weight, ws + wt = 1). To produce the least variable estimate of slant, the visual system must weight each cue by its relative reliability. The reliability of texture information increases with surface slant but does not change with viewing distance. The reliability of disparity information, on the other hand, varies little with slant but decreases with increasing viewing distance. Does the visual system take account of these variations in reliability when estimating surface slant from texture and disparity? To test this we first measured the reliabilities of each cue in isolation across 11 base slants and 3 viewing distances. We used the measured single - cue reliabilities to predict (1) apparent slant when disparity - and texture - specified slant were in conflict and (2) threshold discrimination performance when both cues were available. The apparent slant of cue - conflict stimuli was well predicted by the weighted combination model. We also saw improvements in discrimination when both cues were present. These discrimination thresholds were predicted well for most subjects.