October 2020
Volume 20, Issue 11
Open Access
Vision Sciences Society Annual Meeting Abstract  |   October 2020
Spatial Frequency Filtering: Choices Matter
Author Affiliations & Notes
  • Dirk B. Walther
    University of Toronto
    Samsung Artificial Intelligence Center Toronto
  • Sabrina Perfetto
    University of Toronto
  • John Wilder
    University of Toronto
  • Footnotes
    Acknowledgements  This work was supported by an NSERC Discovery Grant (#498390) and the Canadian Foundation for Innovation (#32896) to DBW.
Journal of Vision October 2020, Vol.20, 1205. doi:https://doi.org/10.1167/jov.20.11.1205
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Dirk B. Walther, Sabrina Perfetto, John Wilder; Spatial Frequency Filtering: Choices Matter. Journal of Vision 2020;20(11):1205. doi: https://doi.org/10.1167/jov.20.11.1205.

      Download citation file:


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

      ×
  • Supplements
Abstract

Spatial frequency content is an important property of natural scenes and has been studied at length. To isolate the effects of specific spatial frequencies, images are typically filtered to include only a particular range of frequencies. The filtered images are then used in experimental tasks assessing observers’ performance in scene content-related task. In addition to selecting the filtering frequency, several other choices that are made in the process of generating filtered images are likely to affect visual perception. We investigated three such choices: the shape of the frequency filter, contrast normalization, and contrast polarity of high spatial frequency-filtered images. We generated filtered images with different settings of these parameters and used them as stimuli in a speeded six-alternative forced-choice scene categorization task. We found that each of the choices strongly affected the categorization performance for low and high pass-filtered images, sometimes determining which of the two resulted in better performance. Filter shape needs to balance a clean frequency cut-off with minimizing ringing artifacts in the filtered image. We recommend using a second-order Butterworth filter as a reasonable compromise. Because of the power spectrum of natural images, filtering images without normalizing contrast severely hurts high pass-filtered images, making them hard to recognize simply due to a lack of contrast. We therefore recommend adjusting contrast energy to be the same for low and high-pass filtered images. High pass-filtered images can be depicted as white lines or black lines on a neutral background without affecting their spatial frequency content. We find that black lines are easier to recognize, presumably due to the cultural habit of drawing by making dark markings on a bright medium. To summarize, we find that several choices in the process of generating frequency-filtered images matter for visual perception. We have quantified these effects empirically and derived specific recommendations.

×
×

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.

×