Abstract
Studies investigating cue-combinations between binocular disparity and shading tend to adopt either forced-choice or method-of-adjustment paradigms. The results of such studies are constrained by the limitations and demand characteristics of each approach. For example, forced-choice paradigms do not allow perceptual biases to be explored; method of adjustment approaches use probes which require prolonged response times, and such probes are problematic for exploring single-cue use.
Visual search was used to explore depth perception from shading, size and binocular disparity. We presented naive observers with a shaded rectangular box, lit from above, containing a circular array of grey discs. For each disc we manipulate the disparity, size and shade independently, as if each cue were drawn from a disc at a different depth. Observers were asked to press a single response button as quickly as possible once they had decided which disc was the ‘deepest’ item in the scene. The scene was then removed and masked. They were then asked to indicate the location of the deepest disc.
We modelled disc selection using a variety of models that weighted and combined the individual cues to depth. Observer selections were not reliably predicted by maximum-likelihood estimates (MLE) of cue combinations (based upon JNDs derived from separate discrimination trials). Rather, deepest disc selections were biased towards disparity cues, even when other discs might be deeper on shade and size axes. Mean performance for MLE model predictions of disc selections were 51% (up to 58%) correct (chance levels were 10-25% depending upon the number of discs in a stimulus). Alternative models, where cue weightings were divergent from channel reliability, achieved 67% correct on average. We conclude that disparity, shading and size cues are not combined as a function of their recent perceptual reliability or discriminability.
Meeting abstract presented at VSS 2012