Item

Abstract

Recent work indicates that the patterns of scales and orientations in an image ('orientation fields') may play an important role in 3-D shape perception. If true, it should be possible to elicit percepts of specific 3-D shapes using stimuli containing only an appropriate orientation field. To investigate this, we use line-integral convolution to 'smear (coerce) 2-D noise patterns to have a geometrically 'correct' orientation field (ie, as similar as possible to the orientation field present in the rendering of some object). The spatial scales are also modulated to be physically accurate. Importantly, the image generation process is entirely based on 2-D filtering operations, and is fundamentally different from a physically realistic rendering. Despite this, the resulting pattern elicits almost as vivid an impression of the object's 3-D shape as a true rendering. Such images were used in depth comparison and gauge figure tasks to assess the relative contribution of spatial scales and orientation fields to shape perception; as well as the dramatic breakdown of shape perception resulting from physically unrealistic orientation and scale configurations. By examining the accuracy with which subjects are able to make metric judgments (relative depth and surface orientation) about the shape percepts evoked by these patterns, we demonstrate that the continuous variations of orientation and spatial scale across an image can play a key role in 3-D shape inference.