September 2011
Volume 11, Issue 11
Vision Sciences Society Annual Meeting Abstract  |   September 2011
What determines the canonical view of a scene?
Author Affiliations
  • Krista Ehinger
    Department of Brain & Cognitive Sciences, Massachusetts Institute of Technology, USA
  • Aude Oliva
    Department of Brain & Cognitive Sciences, Massachusetts Institute of Technology, USA
Journal of Vision September 2011, Vol.11, 850. doi:
  • Views
  • Share
  • Tools
    • Alerts
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Krista Ehinger, Aude Oliva; What determines the canonical view of a scene?. Journal of Vision 2011;11(11):850.

      Download citation file:

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

  • Supplements

Although familiar objects can be recognized from any angle or orientation, there seems to be a preferred viewing angle for most objects. This view, called the “canonical” view (Palmer, Rosch, & Chase, 1981), is the view produced when people are asked to imagine or photograph an object. Are there canonical views of scenes? To answer this question, we conducted an online experiment using 624 panoramic images, which capture the full 360-degrees view around a single location. Observers used an interactive viewer to explore these panoramic scenes and select the “best view” of each location.

We found that agreement on the “best view” of these scenes was generally high. Agreement was higher in indoor scenes than outdoor, probably due to differences in the shape of the scene (agreement was higher in scenes with smaller overall volume and in scenes with a range of visible depths). View agreement was also correlated with name agreement (agreement was higher in scenes with unambiguous names than in scenes which elicited multiple names).

A “volume map” was created for each scene which measured the percentage of the total scene visible in each direction around the camera. We also created a “navigational map” based on the locations of navigational paths (eg, sidewalks, paths between furniture). We tested how well these maps predicted the views chosen by observers (as measured by the area under the ROC curve, AUC). We found that the volume map best predicted the views selected by observers (AUC = 0.75), and although the navigational map performed above chance (AUC = 0.62), it provided no independent predictive power over the volume map. This suggests that the “best view” of a scene is the view which maximizes the amount of visible space, not necessarily a view based on functional constraints such as navigational paths.

Funded by NSF CAREER award to A.O. (IIS 0546262). K.A.E. is supported by a NSF Graduate Research Fellowship. 

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.