Measuring and Modeling the Misinterpretation of 3-D Perspective Views

Abstract

Last year, we showed that there are systematic distortions in the perceptual reconstruction of scenes shown in 3–D perspective views (Smallman, St. John, and Cowen, 2002). We posited a new Cross-Scaling model which states that the depth cue of linear perspective, which is only relevant to widths across the scene (X), is inappropriately used to scale depths into the scene (Y), resulting in systematic misperceptions. Here, we report three experiments that test and refine this Cross-Scaling model and we relate it to a broader array of tasks and phenomena. In Exp 1, participants reproduced a square grid as it would appear to them from an oblique perspective. In Exp 2, they picked a drawing, from 25 alternatives, that closest approximated the Exp 1 scene. In Exp 3, participants completed square grids on the ground of 3-D perspective views of scenes that contained only linear perspective or foreshortening depth cues. In Exps 1 & 3, in accordance with the model, participants consistently made grids in which X and Y tapered with distance (non-veridically) equally. In Exp 2, however, participants chose as most natural, an incorrect drawing, but one other than that predicted by the model. Overall, there is apparently a general misconception among viewers about the nature of perspective projection, similar to Naive Physics misconceptions, that accounts for a wide variety of previously unrelated perceptual phenomena; these have important human factors implications.