An effect of referential scene constraint on search implies scene segmentation

Abstract

Subjects searched aerial images for a UFO target, which appeared hovering over one of five scene regions: Water, fields, foliage, roads, or buildings. Prior to search scene onset, subjects were either told the scene region where the target could be found (specified condition) or not (unspecified condition). Search times were faster and fewer eye movements were needed to acquire targets when the target region was specified. Subjects also distributed their fixations disproportionately in this region and tended to fixate the cued region sooner. We interpret these patterns as evidence for the use of referential scene constraints to partially confine search to a specified scene region. Importantly, this constraint cannot be due to learned associations between the scene and its regions, as these spatial relationships were unpredictable. These findings require the modification of existing theories of scene constraint to include segmentation processes that can rapidly bias search to cued regions.

Keywords:

• Visual search
• Scene context
• Aerial images
• Region segmentation
• Eye movements

Acknowledgements

This work was supported by the National Science Foundation under Grant IIS-0527585 and by the National Institute of Mental Health under Grant 2-R01-MH63748 to GJZ. We thank Zainab Karimjee, Alyssa Fasano, Rosa Vespia, and Mariya Norenberg for help with stimulus creation and data collection.

Notes

¹Of course many studies have also used spatial precues to manipulate the expected location of a target (e.g., Palmer, Ames, & Lindsey, 1993; Zelinsky, 1999), but these explicit spatial cueing situations are not the focus of this paper.

²We use the term “target region” to refer to the collective instances in a given scene of the region type containing the target, not the specific region in which the target is actually located. For example, if the saucer is located over a building, all fixations on any building are considered fixations in the target region.

³So as to minimize biases arising from initial central fixation and target inspection during discrimination, this analysis starts with the fixation following the initial saccade and ends with the first fixation within 1.5° of the target.