Abstract
The contextual scene in which an object appears can provide critical information for identifying that object. In previous studies, Barenholtz (VSS, 2009) found that participants could identify objects on the basis of reduced visual information when those objects appeared in context than when they appeared in isolation. The current study assessed the contribution of different sources of information available in a contextual scene. We compared performance – as measured by the level of resolution needed for recognition – across five contextual preview conditions: 1) General Scene, in which subjects first saw a preview of the contextual scene in which the object had originally appeared (the target object was digitally removed from the image in all of the preview scenes); 2) Object Location, in which the preview of the scene also included a cue to the location of the object; 3) Object Size, in which the preview was followed by a size referent; 4) Object Location and Size which combined conditions 2 and 3 and 5) No Context, in which there was no scene preview. We considered each of these different contextual conditions for both ‘Experts’ (where the scene stimuli were drawn from photographs of their own homes) vs. ‘Novices’ participants, who were not familiar with the scenes. Some of the major findings included an effect of expertise across all context conditions; Experts, but not Novices, showed significant benefits from location information; a strong deficit in the no-context condition for both Expertise groups. In additional analyses, we included independently obtained ratings of the target objects along several dimensions including the consistency of the object type within the scene category, the frequency of such object-types in those scenes and the typicality of the particular object as an example of that object-type. In general, higher consistency ratings were most strongly correlated with improved performance.
National Science Foundation.