Abstract
A fundamental problem in the study of human vision is how information accumulates across separate fixations into an overall representation of the scene. There is a surprising lack of consensus on the basic properties of scene memory. We have studied the build up of memory for objects in scenes (12/scene), in a task requiring eye movements to scan the display. Memory performance improved over trial duration, and improved further when displays were re-tested on subsequent trials. The number of items recalled after long trials was nearly equivalent to performance after a sequence of short trials separated in time by several minutes. Memory build up was affected by changes in scene context. Eye movement strategy depended on trial length, but subjects remembered both fixated and non-fixated objects. These results suggest that memory for objects in a scene is better than previously thought.