The probabilistic foundation of visual space


Journal Article

An assumption in many studies is that visual space (i.e., the space we perceive) is metrical. For example, perceived space has often been considered a Riemann space of constant curvature. In such cases, perceived spatial relationships should be independent of the context of the visual scene. This category of assumptions, however, is inconsistent with numerous experimental observations showing that the relationship between the perceived and the physical parameters of scene geometry is systematically distorted. In the absence of a principled account of what this distortion of physical space actually means, other investigators have assumed that visual space is either affine or subject to some other transformation of physical space. Here we have explored an alternative hypothesis, namely that visual space is generated solely by the statistical properties of the physical world. To this end, we acquired and analyzed a database of natural scenes in which the distances of all object points from the image plane were measured with a laser range scanner. The probability distributions of these distances are scale invariant, a feature that accords with the human perception of distance and location under impoverished stimulus conditions. Furthermore, the probability distributions of the physical sources of visual stimuli (i.e., their distance, depth, size, and surface orientation) were found to be systematically influenced by the range distribution of the surround. These context-dependent probability distributions of physical sources generally account for the known distortion in the perception of distance, depth, size, and orientation (e.g., the "terrain influence" on distance judgment, and the well-known contextual effects that influence the perception of orientation). Our results thus suggest that visual perceptual space, for reasons of biological advantage, is straightforwardly determined by the probability distributions of the sources underlying visual stimuli.

Full Text

Duke Authors

Cited Authors

  • Yang, Z; Purves, D

Published Date

  • December 1, 2002

Published In

Volume / Issue

  • 2 / 7

International Standard Serial Number (ISSN)

  • 1534-7362

Digital Object Identifier (DOI)

  • 10.1167/2.7.715

Citation Source

  • Scopus