Fig. 2

Possible projection models for viewers’ implicit assumptions in shape perception. (a) Conventional linear perspective uses a single COP for the entire picture. The original vantage-point compensation hypothesis²⁴ states that viewers interpret pictures according to an estimate of the picture’s true COP, but extensive experimental evidence—combined with the fact that many pictures do not have COPs—contradict this hypothesis^22,23. As depicted in the inset, linear perspective causes objects in the periphery to appear distorted, a phenomenon known as marginal distortion. (b) In the Direct View Condition^19,30 hypothesis, image regions appear undistorted when they are consistent with a local-linear perspective that has the COP in front of the object. The inset shows a picture for which the region containing the green object is consistent with a local-linear perspective with effective COP in front of the object. Therefore, in this case, it does not appear distorted..