Abstract
We recently reported the existence of physically different stimuli, containing different patterns of disparity, that are perceptually indistinguishable from one another (ARVO, 2001). In one stimulus (the “natural” stimulus), the disparity and eye position signals were correct for a frontoparallel surface. In the second stimulus there was a cue conflict: two stereoscopic methods of estimating slant (one based on vertical disparity, the other based on felt eye position) yielded slant estimates of opposing sign (Gillam, Chambers, & Lawergren, 1988; Banks & Backus, 1998). The natural and conflict stimuli both evoked the same near-zero apparent slant. In addition, the stimuli were indistinguishable from one another in a psychophysical task. Thus, they were true perceptual metamers. Furthermore, they could be made distinguishable by a change of vergence posture alone, without changes to the retinal images. Here, we show that the estimator reliability theory of Backus & Banks (1999) predicts these results. First, the cues available to construct a visual percept were formally described as distinct estimators of some scene parameter (such as surface slant). These estimators were then combined in a weighted average, using weights proportional to the estimators' reliabilities (Landy, Maloney, Johnston & Young, 1995). We assumed that the visual system knew the reliabilities of the estimates, and was able to update the weights as reliabilities changed across viewing conditions. How the estimators' reliabilities changed across viewing conditions was determined from a Monte Carlo simulation that assumed fixed Gaussian error in the measurement of the signals used by the estimators. For the stimuli in these experiments, the theory predicted that as vergence decreased, relatively more weight would be given to the slant estimate based felt eye position. This correctly predicted the sign of the change in perceived slant that occurred with the change in vergence.