Abstract
There are many known instances in which the visual system combines redundant cues (or estimates) by weighted averaging, with weights determined by the relative reliabilities of the constituent estimators. There are also documented instances in which thresholds for multiple-cue stimuli are reduced by an amount consistent with cue weighting: the reliability (reciprocal variance) of an estimate based on independent estimators was found to be roughly optimal—equal to the sum of the estimators′ reliabilities (e.g. Backus & Banks, 2000). We tested whether these principles characterize depth perception for two stereoscopic depth cues. The relative disparity between two points in a scene is determined by their positions relative to the head. Retinal relative disparity (RRD) is a well-known signal caused by depth. Delta-vergence (DV) is the change in vergence required to fixate each of two targets in turn; it is equal to their relative disparity. Both RRD and DV can be measured by the visual system, and both are used during depth perception. Their relative weights are a function of target separation: DV is used to a greater extent when targets are widely separated (Backus & Matza-Brown, 2003). We measured observers′ thresholds for RRD and DV separately, and in a stimulus containing both cues. The relative disparity threshold was lower for the combined-cue stimulus than for either cue alone, consistent with theory. The relative weighting given by the visual system to RRD and DV was measured using cue conflict stimuli; it was also consistent with theory. At each target separation, the empirically measured weights varied according to the relative reliabilities of the two estimators, as calculated from the reciprocal squares of their respective relative disparity thresholds.
EY013988