Influences of context on object detection and identification in natural scenes

Abstract

The way we perceive complex visual scenes has been an area of much research and debate. Many studies have found that the context of a scene is used to guide attention to important and relevant areas of a scene. Other studies, however, have found that objects that are incongruent with the scene context capture attention. These contradictory findings have been found both within and across tasks. The purpose of the present research was to reconcile these contradictory results. Two processes were identified as underlying complex scene perception: object detection and object identification. Further, the current research demonstrates the relative weighting of these processes differs according to task demands; some tasks weight object detection more heavily, whereas other tasks weight object identification more heavily. Moreover, it was demonstrated that the weighting of these processes can be manipulated within a task in such a way as to produce either congruency benefits or costs. Finally, in circumstances in which processing is weighted in favour of object detection, it was demonstrated that eye gaze, and presumably overt attention, is captured by semantically incongruent objects early into scene perception. The current research helps our understanding of complex scene perception by reconciling contradictory findings reported in previous studies. In particular, two processes were identified: object identification, which relies on a congruent context, and object detection, which relies on an incongruent context. In this way, past experience may promote efficient scene perception by promoting the use of regularities in the environment (e.g., congruent context), but also leaving the attention system sensitive to areas of the scene that contradict the expectations set by the context (e.g., incongruent objects).