Abstract
A convergent set of findings suggests that the phenomenal impression of a scene's structure contains representations of some of the viewed objects, set at relatively precise spatial locations in the visual field. The representation is thought to be limited to 4–5 objects by the capacity of visual short-term memory (Hunter, Warlaumont, and Edelman, 2005; Irwin and Zelinsky, 2002). Yet, our ability to rapidly extract the gist of real-world scenes or decide which objects are congruent with scene context argues for much greater capacity (Davenport and Potter, 2004; Rousselet, Joubert, and Fabre-Thorpe, 2005). What accounts for the disparity in these findings? We attempt to bridge the gap between scene research using naturalistic photographs and that using artificial scenes of parameterized stimuli by investigating what aspects of scenes increase subjects' ability to remember spatial positions of objects. Subjects were shown scenes composed of photographs of diatoms randomly located against 1) no background; 2) a background of rippled sand; or 3) sand with two fixed “landmarks,” for 6 s, with one scene intervening between sample and test. Subjects indicated whether or not a designated object was in the same location as in any previous scene. Trial blocks with landmarks showed significantly better performance than blocks with sand but no landmarks; both were better than blocks with no background. Scene backgrounds and landmarks may work to activate brain areas that represent spatial geometry (Epstein, 2005), helping to “anchor” object locations and thereby boosting visual memory capacity.