Abstract
The visual integration of different cues for perceiving a three-dimensional (3D) shape is currently understood in terms of a two-stage model hypothesizing that (i) a 3D estimate is derived separately from each cue, and (ii) a weighted combination of these estimates is computed through the estimated reliabilities of the cues (Landy, Maloney, Johnston & Young, 1991). Interactions among cues are allowed only for estimating missing parameters necessary for a 3D interpretation of each cue in isolation. Domini and Caudek (2004) hypothesized, conversely, that perception of 3D shape relies on the natural covariance among the retinal signals defining different visual cues, and does not require separate 3D estimates. In the present experiment we investigate the effects of adaptation to a relation between stereo and motion signals violationg the constraints of natural optics. The observers participated to 4 sessions in 2 consecutive days. In all phases of the experiment, the stimuli were defined by stereo and motion information. Each session was made up of a pre-test, an adaptation phase, and a post-test. In both the pre- and the post-test, participants were required to discriminate between concave/convex random-dot rigid surfaces. In the adaptation phase, participants were required to discriminate between random-dot volumes with different depth-extents. In the pre- and post-test, the stimuli preserved the natural linear relation between disparity and velocity signals. In the adaptation phase, the relation between disparity and velocity signals was non-linear. We found a shift in the psychometric functions of the post-test, but only in the second day of testing (after a night of sleep). These results (i) cannot be explained in terms of adaptation to 3D shape (e.g., Poom & Borjesson, 1999), (ii) are inconsistent with a modular theory, and (iii) are predicted by the Intrinsic Constraint model (Domini & Caudek, 2004).
Supported by NSF grant BCS 0345763