Abstract
OBJECTIVE: Surface transparency can result in the mixing of image textures projecting from multiple surfaces in the scene. At large viewing distances, the inference of multiple surfaces depends primarily on the identification of independent texture distortions due to surface attitude and shape. In prior work (VSS 2001), we showed that two overlaid transparent textured surfaces differing by more than 90 deg in tilt could be reliably identified. Further, when both surfaces were seen, estimation of tilt was unaffected by the other textured surface. For smaller tilt differences, when only a single surface was seen, the estimated tilt could be predicted based upon an optimal probabilistic fusion model. In the present study, we determine whether these results extend to the perception of surface slant.
METHODS: Participants monocularly viewed two transparent textured planar surfaces rendered in perspective projection within a 24 deg diameter window. The mean slant of the two surfaces was random and uniformly distributed over [30,40] deg. The slant difference between the two surfaces was systematically varied over [−40,40] deg. Participants were asked to a) indicate whether one or two surfaces were seen and b) estimate, using a mouse-controlled gauge figure, the 3D attitude(s) of the perceived surface(s).
RESULTS: Our results for slant estimation differ dramatically from those for tilt estimation. When participants were able to see two surfaces, the presence of the second surface was found to decrease the reliability of slant estimation. When only one surface was seen, perceived attitude was dominated by the surface of greater slant. This bias cannot be predicted from measured uncertainties in estimating the slants of the individual surfaces.
CONCLUSION: The perception of overlapping transparent surfaces viewed at a distance is strongly affected by a “slant capture” effect that cannot be predicted by optimal estimation models.