Abstract
Most studies of visual cue integration have found that observers' data can be modeled by a linear cue combination rule. We conducted an experiment in which, intuitively, a linear cue combination rule should not provide a satisfying model of observers' data. The experiment used a surface defined by texture and shading cues. At a coarse-level of detail, the surface was planar, but it also contained deformations resembling random bumps or waves. Subjects performed a slant discrimination task—they judged whether comparison surfaces were more or less slanted than a standard surface—in three conditions, namely when surfaces were defined by the texture cue alone, by the shading cue alone, or by both cues. The results indicate that when both cues were available subjects' discrimination thresholds were significantly lower than predicted by a linear model. We propose that the texture and shading cues to slant are each highly ambiguous when there is great uncertainty about the shape of the surface. When only a single cue is available, observers find it difficult to estimate surface shape (due to its irregularities) and, thus, find it difficult to interpret that cue for the purpose of judging slant. When both cues are available, however, estimates of surface shape are more accurate, thereby making it easier to interpret the cues as indicators of slant. The results are consistent with the hypothesis that observers estimate surface shape, and use this estimate when judging surface slant.
This work was supported by NIH research grant RO1-EY13149.