Abstract
The 3D orientation of planar surfaces can be perceived in the absence of edge information. Here, we provide a theoretical and empirical demonstration that 3D orientation is specified by the orientation disparity of surface contours (i.e., intrinsic lines belonging to the same plane).
The Local Orientation Disparity of correspondent lines [LOD] was described as a function of the Local Average Orientation [LAO], taking egocentric distance and surface inclination relative to the vertical and to the horizontal as unknowns. We found that unknowns are recoverable by fitting such a function to the orientation disparity map (i.e., to the set of points in LOD-LAO space used as a compact representation of the impoverished stimulus).
We tested the model by instructing observers to set the V/H-inclination of a planar surface specified by randomly-oriented lines intersecting in a common point. The surface was visible through a circular aperture in a fronto-parallel screen. Observers produced 64 surface orientations resulting from the factorial combination of 4 repetitions by 16 V/H-inclinations in the 0–45 deg range. Despite the absence of matchable points and the high portion of un-matchable regions, estimated surface orientation was highly correlated with V/H-inclination. Systematic orientation biases were also observed.
Empirical results, including bias effects, highly correlated with model predictions, suggesting that the orientation-from-disparity problem can be solved by matching input data to the implicit knowledge on the LOD-LAD correlation. An ideal observer capable of encoding orientation disparities can achieve a veridical 3D representation by performing the mathematical analysis described in our model.