September 2018
Volume 18, Issue 10
Open Access
Vision Sciences Society Annual Meeting Abstract  |   September 2018
Human Object Detection in Natural Scenes: Evidence From a New Dot Probe Task
Author Affiliations
  • Colin Flowers
    Department of Psychology, University of Arizona
  • Mary Peterson
    Department of Psychology, University of ArizonaCognitive Science Program, University of Arizona
Journal of Vision September 2018, Vol.18, 393. doi:https://doi.org/10.1167/18.10.393
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Colin Flowers, Mary Peterson; Human Object Detection in Natural Scenes: Evidence From a New Dot Probe Task. Journal of Vision 2018;18(10):393. https://doi.org/10.1167/18.10.393.

      Download citation file:


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

      ×
  • Supplements
Abstract

Object detection is often measured by assessing whether an object of a given category was present in a briefly exposed scene. With this task, some accurate responses might be based upon scene context. Here, we measure object detection by instructing participants (n = 53) to report whether a colored flashing dot probe located near a border in a colored photograph of a natural scene was "on" or "off" the object bounded by that border (100-ms masked exposure). Dot location varied from central to peripheral regions of the photographs. Accurate responses were taken to index accurate object detection. This method reveals object location rather than category. We investigated category effects by presenting 741 photographs from the Common Objects in Context (CoCo) set containing objects in 10 different categories (54 – 89 objects per category). Overall accuracy was 65.34%. A one-way ANOVA showed a significant effect of category (p < 0.001). Performance was poorest (although significantly greater than chance) on small textureless objects (e.g. knives/forks; 57.11%), or objects that were often occluded in the photographs (e.g., bowls/cars; 58.26%). These objects often caused computer vision models of object detection to fail. Detection accuracy for these objects was significantly lower than all other categories, ps < 0.04. Performance was best on textured objects that were generally the focal point of the photographs (e.g., airplanes/birds/zebras - 74.61%, were detected significantly better than objects in all other categories, ps < 0.03), objects that are often accurately detected by computer vision models. Performance with the objects in the other categories (apples/chairs/people) was intermediate. Our method, emphasizing locating objects within scenes, provides evidence regarding human object detection while eliminating guessing based on context. We plan to use black and white photographs to increase the visibility of the colored flashing dot probe to determine whether this pattern remains unchanged.

Meeting abstract presented at VSS 2018

×
×

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.

×