Purchase this article with an account.
Paul A. Warren, Pascal Mamassian; The dependence of slant perception on texture orientation statistics. Journal of Vision 2003;3(9):847. doi: 10.1167/3.9.847.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
PURPOSE: Many studies have addressed the problem of extracting surface orientation from texture (e.g. Blake et al. 1993; Buckley et al. 1996; Knill, 1998; Li & Zaidi, 2001). Few, however, have looked specifically at the information available from the first and second moments of the orientation distribution of linear texels. There is evidence that observers are able to extract this information (Dakin & Watt, 1997) and use it to recover slant and tilt under orthographic projection (Witkin, 1981). Here we assess the extent to which human slant perception is biased by texture orientation statistics. METHODS: Observers briefly saw patches of texture through a circular aperture subtending approximately 20 degrees visual angle. The texture comprised of small line segments on a planar surface. Texel orientation on the surface followed either a uniform distribution (the reference) or a unimodal circular distribution with variable bias and spread parameters (the comparison). The plane was rotated about a horizontal axis and seen under perspective projection. In each trial observers saw a reference and comparison surface and were asked to indicate which had larger slant (top end away). Predictions were obtained by computing the closed form likelihood function describing the probability that a surface slant produces a particular image under perspective projection. RESULTS: Texture orientation statistics had a significant effect on perceived slant. Observers overestimated slant relative to the isotropic case when texel orientations were biased towards the horizontal, and inversely underestimated slant when they were biased vertically. CONCLUSIONS: Human observers rely on texture orientation statistics to recover surface orientation. Our data are consistent with a Bayesian model which uses the assumption of isotropy. This model has a uniform prior for surface orientation and a decision rule which selects the expected value of the posterior distribution.
This PDF is available to Subscribers Only