August 2014
Volume 14, Issue 10
Free
Vision Sciences Society Annual Meeting Abstract  |   August 2014
Border salience reveals a curved global geometry of the perceptual space of local image statistics
Author Affiliations
  • Syed M. Rizvi
    Brain and Mind Research Institute-Weill Cornell Medical College
  • Mary M. Conte
    Brain and Mind Research Institute-Weill Cornell Medical College
  • Jonathan D. Victor
    Brain and Mind Research Institute-Weill Cornell Medical College
Journal of Vision August 2014, Vol.14, 654. doi:https://doi.org/10.1167/14.10.654
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Syed M. Rizvi, Mary M. Conte, Jonathan D. Victor; Border salience reveals a curved global geometry of the perceptual space of local image statistics. Journal of Vision 2014;14(10):654. https://doi.org/10.1167/14.10.654.

      Download citation file:


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

      ×
  • Supplements
Abstract

Local features such as lines, edges, and corners are the elements of form vision. Segmenting an image and estimating its surface properties depend on analyzing the spatial statistics of these components. Thus, visual textures constructed from these features constitute an important "perceptual space," in which distances correspond to the degree to which a texture difference supports the inference of a border. Here, we characterize basic aspects of the global geometry of this perceptual space. To reduce the dimensionality of the space to a practical level, we consider binary images and parameterize them by the configurations present in 2x2 neighborhoods, a strategy that focuses on image statistics that are informative in natural images (Tkačik et al., 2010). This leads to a 10-dimensional space of synthetic images, which constitutes a perceptual space, analogous to the familiar 3-dimensional color space. Our previous discrimination-threshold measurements (Victor et al., VSS 2013) showed that perceptual distances near the origin of this space corresponds to a Euclidean metric, and hence, a locally flat geometry. Here, we use suprathreshold border salience judgments to probe the global geometry of this space. Subjects (N=4) were presented with a brief (120 ms) four-quadrant display, each containing a texture sample. Texture samples were chosen to represent points on a line in the space of image statistics. Subjects judged which of the four texture borders was most salient. The relative salience judgments were then subjected to multidimensional scaling. In some directions, the inferred perceptual geometry matched the linear sampling of the space. But in other directions, the linear sampling of the space translated to a perceptual geometry in which opposite edges of the space were warped inward to meet each other. These findings constrain models for how the perceptual space is constructed, and, in particular, imply that opponent mechanisms alone do not suffice.

Meeting abstract presented at VSS 2014

×
×

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.

×