July 2013
Volume 13, Issue 9
Vision Sciences Society Annual Meeting Abstract  |   July 2013
Statistical coding of natural closed contours
Author Affiliations
  • Ingo Fründ
    Centre for Vision Research, York University, Toronto, ON, Canada
  • James Elder
    Centre for Vision Research, York University, Toronto, ON, Canada
Journal of Vision July 2013, Vol.13, 119. doi:https://doi.org/10.1167/13.9.119
  • Views
  • Share
  • Tools
    • Alerts
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Ingo Fründ, James Elder; Statistical coding of natural closed contours. Journal of Vision 2013;13(9):119. doi: https://doi.org/10.1167/13.9.119.

      Download citation file:

      © ARVO (1962-2015); The Authors (2016-present)

  • Supplements

We seek to understand the statistical regularities in the bounding contours of natural shapes, and how the human visual system exploits these regularities for perceptual grouping and object recognition. Here we employed a dataset of 391 animal shapes, approximated as equilateral polygons. From this dataset we extracted a set of low-order statistical features, including expected circular variance, skew and kurtosis of angles as well as the circular correlation between neighbouring angles.

To measure human selectivity for these features, we developed a method for generating contour metamers that match the natural contours on selected subsets of these features, but which lack all other statistical regularities found in the natural shapes. The main challenge here is the constraint that metamers be simple (non-intersecting) closed contours. To solve this problem, we developed a novel method for constructing and sampling from generative maximum entropy models that satisfy all of these constraints.

Psychophysical Methods. In a two-interval task without feedback, observers were asked to distinguish between two fragments of contour, one from an animal shape and one from a metamer. We measured the length of the contour fragment required for threshold performance.

Results. Matching the expected variance between animal and metamer shapes raised thresholds for all observers, pointing to human selectivity for angular variance information (variation in curvature). At the same time, an ideal observer using only the expected variance performed much better than humans at discriminating unconstrained metamers and animal shapes, indicating that human encoding of this cue is far from perfect.

Interestingly, ideal observer models that also exploit higher-order cues (expected skew, kurtosis, correlation between neighbouring angles) fall short of human performance for discriminating natural shapes from metamers that are matched in angle variance, suggesting that humans rely on more complex global cues for discriminating natural shapes.

Meeting abstract presented at VSS 2013


This PDF is available to Subscribers Only

Sign in or purchase a subscription to access this content. ×

You must be signed into an individual account to use this feature.