Abstract
In the Bayesian approach to perception (Knill & Richards, 1996), prior assumptions play a critical role. However, little is known about how the brain learns and adapts its “priors”. It has been proposed that priors reflect environmental statistics (Geisler & Diehl, 2003). As the statistics of different environments, for example man-made vs. natural environments, vary strongly, it would be advantageous to have different priors for different environments and flexibly switch between them. To test this, we had subjects judge the 3D-orientation of planar objects containing small conflicts between the slant suggested by binocular disparities and figural compression. The informativeness of the latter, monocular cue depends on subjects' priors, because inferences from two-dimensional information to a three-dimensional interpretation require assumptions about symmetry, isotropy, and such. Consequently, when we manipulated the statistical regularities of the environment in which the stimuli were presented, subjects' reliance on the compression cue changed. In a first session, the environment in which test stimuli were presented consisted of isotropic objects, and in this regular environment, subjects relied about equally on the binocular disparity and the compression cue to slant. In the following sessions, we randomly interleaved trials with regular environments and trials with irregular environments which consisted of non-isotropic objects with random aspect ratios. The influence of the compression cue was stronger in trials with regular than in trials with irregular environment, and in both cases significantly lower than in the first session in which only regular environments were presented. This indicates that subjects generally lowered their belief in isotropy when trials with irregular environments were introduced and used a stronger prior on isotropy in trials with regular than with irregular environments. We conclude that subjects can rapidly switch between priors used to interpret a monocular cue to depth when shifting between environments with different shape statistics.
DFG TR528 1-3 to A.S. and J.T. NIH R01-EY13319 to D.K.