Ruth Rosenholtz¹; ¹NVIDIA Research

Many researchers have studied the human ability to perceive 3D shape and depth from a single 2D picture. How accurate are these judgments, what cues do they rely on, and what ambiguities are inherent to a given cue? Why are people often insensitive to viewing a picture from the wrong center of projection, and when do they perceive distortions? Such 3D perception depends upon lower-level judgments that often involve comparisons across space. For example, utilizing simple perspective cues requires comparing the angle between two separated lines, parallel in the world but converging in the image. If we understood how well humans can make such comparisons, that would inform our understanding of the accuracy and ambiguities inherent in picture perception. Two experiments measured orientation discrimination across space, both for a pair of isolated lines and for two lines forming opposite sides of a quadrilateral. Ten observers were asked to judge which line was tilted more clockwise. The lines had an average length of 2.9 deg, and the spacing between them was 0.95, 1.9, 3.8, or 7.6 deg of visual angle. On each trial, a staircase determined the difference in orientation, with the base orientation either near-horizontal or oblique. Performance degraded significantly as a function of distance between the lines, and was worse when the lines formed part of a quadrilateral than when isolated. In addition, we found a typical oblique effect. In the easiest condition (isolated lines, near horizontal, smallest spacing) thresholds were comparable to prior work (1.2 deg). However, in the hardest condition (quadrilateral, oblique, largest spacing), thresholds rose to approximately 6.9 deg. This may strongly constrain our ability to perceive shape from perspective and notice distortions.