December 2022
Volume 22, Issue 14
Open Access
Vision Sciences Society Annual Meeting Abstract  |   December 2022
Subtle differences in the perceptual spaces of low-level features and objects
Author Affiliations & Notes
  • Suniyya A. Waraich
    Weill Cornell Graduate School of Medical Sciences
  • Jonathan D. Victor
    Weill Cornell Medical College
  • Footnotes
    Acknowledgements  EY 07977
Journal of Vision December 2022, Vol.22, 3222. doi:
  • Views
  • Share
  • Tools
    • Alerts
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Suniyya A. Waraich, Jonathan D. Victor; Subtle differences in the perceptual spaces of low-level features and objects. Journal of Vision 2022;22(14):3222.

      Download citation file:

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

  • Supplements

Sensory and semantic information are qualitatively different: while color lies on a continuous spectrum, objects are often categorical. We hypothesize that these two kinds of domains are represented in spaces that have different geometries. To test this, we characterized the perceptual spaces of five stimulus domains ranging in their degree of semantic content, from textures to animal names. In between, there were animal images, and two intermediate domains comprised of texturized images — one texture-like and difficult to recognize; one image-like and easier to recognize. All stimuli were derived from the same 37 common animals chosen from WordNet. In parallel psychophysical experiments, we gathered similarity judgments between pairs of stimuli for each domain. In each trial, 8 stimuli appeared around a central reference that subjects clicked in order of perceived similarity to the reference. Each trial was repeated 5 times over ten hour-long sessions. Our design included pairs of trials in which the same triplet — a reference and two other stimuli — appeared with different surrounding stimuli, allowing us to probe context effects. We found responses were very consistent across subjects (2M, 7F) and contexts (n=7). Using a variant of multidimensional scaling, we assessed how well Euclidean models explained the similarity data. Whereas all domains required over five dimensions to fully explain the data, hierarchical clustering on the points obtained from the 5D Euclidean models revealed a consistent difference: dendrograms derived from points in word space were more balanced than those from the texture and texture-like spaces. Finally, we validated the analysis with control experiments using domains with known perceptual spaces of 3 dimensions: color and uncorrelated grayscale textures. Thus, the geometric differences between the perceptual spaces of domains along this sensory to semantic gradient are subtle and are manifest by their tree-like organization.


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.