Abstract
Data obtained from slant discrimination studies with textured surfaces (e.g., Knill & Saunders, Vis Res, 2003) have consistently shown that the reliability of observers' slant estimates is higher for large slants than for small slants, but the opposite result has been reported for the variability of slant estimates that are obtained using an adjustment task (Todd, et al, Vis Res, 2005). In order to determine the reliability of slant estimates from discrimination thresholds, it is typically assumed that observers' judgments are veridical except for internal random noise. This assumption is necessary because thresholds can also be affected by any systematic biases in perceived slant. Indeed, the results from prior discrimination studies are perfectly consistent with a changing bias as a function of depicted slant rather than a change in reliability, and this could potentially reconcile the conflicting results that are obtained using matching tasks. In an effort to confirm this possibility, the research described here was designed to measure the psychophysical mapping between physical and perceived slant from texture, under viewing conditions that are similar to those employed in previous studies. Observers were presented on each trial with two pairs of adjacent surfaces with different slants, and they were asked to adjust the slant of one of those surfaces so that the apparent angle between each pair was equated. The results reveal that there is an upward curving psychophysical function between physical and perceived slant, such that a depicted angle between relatively shallow slants appears significantly smaller than the same depicted angle between steeper slants. These findings indicate that the assumptions employed in prior discrimination studies for shape from texture may not be valid, which may also make it necessary to reconsider how discrimination procedures are employed for the investigation of cue combination.
This research was supported by a grant from NSF (BCS-0546107).