Not all boundaries have the same status in vision. Gottesman and Intraub (2003) showed that boundary extension occurs at view-boundaries but not at boundaries within a view. Does this “special status” of view-boundaries influence the representation of occluded objects in scenes? To test this we created a special set of pictures and a new procedure that allowed us to eliminate many of the constraints inherent in typical picture memory tests. Using a multi-layered graphics program, viewers were able to independently adjust view-boundaries, object placement, and amount of occlusion. Twelve multi-object scenes were photographed and two versions of each were created in which a target object was cropped by an object or cropped the same amount by the edge of the view. One version of each scene was shown to 72 viewers for 10s, followed by a 1s visual noise mask. The scene reappeared without the target object and one other object: the borders were pulled out for of the viewers revealing more of the scene than had the stimulus, and were pushed in for the other , revealing only a tiny central portion of the scene. Viewers adjusted the borders, the missing objects reappeared and viewers placed them in their remembered positions. At object-boundaries viewers tended to remember the occlusion well, whereas at view-boundaries they tended to remember having seen 10-12% more of the target than was originally shown. However the visual information at the view-boundary is not the same as that at the object boundary because there is no information outside the view. Experiment 2 (N=72) replicated the results even when an occluding object was “stamped” around the view-boundary so that local information at the point of occlusion was controlled across conditions. Results showed that not all occlusion is the same - view-boundaries encourage extrapolation that may aid integration of views, whereas object-boundaries tend to preserve object-relations within the view.
Research supported by NIMH grant R01 MH54688-04A2