August 2012
Volume 12, Issue 9
Vision Sciences Society Annual Meeting Abstract  |   August 2012
Natural image statistics based population coding for local edge probability
Author Affiliations
  • Chaithanya Ramachandra
    Biomedical Engineering Department, University of Southern California
  • Bartlett Mel
    Biomedical Engineering Department, University of Southern California\nNeuroscience Graduate Program, University of Southern California
Journal of Vision August 2012, Vol.12, 846. doi:
  • Views
  • Share
  • Tools
    • Alerts
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Chaithanya Ramachandra, Bartlett Mel; Natural image statistics based population coding for local edge probability. Journal of Vision 2012;12(9):846.

      Download citation file:

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

  • Supplements

A key computation in the visual cortex is the extraction of object contours, where the first stage in the process is thought to be local edge detection by V1 simple cells. From an engineering perspective, however, individual Gabor-like receptive fields exhibit relatively poor localization and orientation tuning on natural object contours. Higher quality information about object boundaries can in principle be decoded from the pattern of activity over the local population of simple cells in the vicinity of an object edge. To understand how local excitatory and inhibitory interactions among simple cells can best be used to estimate local edge probability in natural scenes, we used Bayes rule to compute P(edge @ x,y,theta|r0,r1...rN) where r0 is the response of the linear filter at (x,y,theta), and r1...rN are the responses of other filters in the vicinity of (x,y,theta). We modeled the joint distribution P(r0,r1...rN|edge {or ~edge} @ x,y,theta) based on (1) measurements of individual filter statistics on and off edges in natural scenes using human ground truth labels, and (2) simplifying assumptions about the dependencies between filter responses in the vicinity of an edge. The resulting "formula" for computing local edge probability, expressed in a 2-layer format (sum of sigmoids of sums), led to significantly improved localization of edges both in space and orientation, improved precision-recall measures compared to the linear filter, and visually improved edge detection results. This also suggests a biologically-plausible scheme for combining neuronal responses through local feedforward excitatory and inhibitory interactions.

Meeting abstract presented at VSS 2012


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.