Annu. Rev. Psychol. 1999. 50:243–71Copyright Ó 1999 by Annual Reviews. All rights reservedHIGH-LEVEL SCENE PERCEPTIONJohn M. Henderson and Andrew HollingworthDepartment of Psychology, Michigan State University, East Lansing, Michigan48824; e-mail: [email protected], [email protected] WORDS: eye movements, vision, scene identification, object identification, changeblindnessABSTRACTThree areas of high-level scene perception research are reviewed. The firstconcerns the role of eye movements in scene perception, focusing on the in-fluence of ongoing cognitive processing on the position and duration of fixa-tions in a scene. The second concerns the nature of the scene representationthat is retained across a saccade and other brief time intervals during ongoingscene perception. Finally, we review research on the relationship betweenscene and object identification, focusing particularly on whether the mean-ing of a scene influences the identification of constituent objects.CONTENTSINTRODUCTION .......................................................... 244EYE MOVEMENT CONTROL IN SCENE PERCEPTION ......................... 245Fixation Position During Scene Perception ................................... 245Fixation Time During Scene Perception ...................................... 251Conclusions .... ........................................................ 252SCENE MEMORY ACROSS SACCADES . . . ................................... 253Change Blindness Across Saccades During Scene Viewing ....................... 254Change Blindness and Simulated Saccades.................................... 255Conclusions .... ........................................................ 258SCENE CONTEXT AND OBJECT IDENTIFICATION............................ 258Scene Identification ...................................................... 258Models of Object Identification in Scenes ..................................... 260Studies of Object Identification in Scenes ..................................... 262Conclusions .... ........................................................ 267CONCLUSION ......................... ................................... 2680084-6570/99/0201-0243$08.00243Annu. Rev. Psychol. 1999.50:243-271. Downloaded from arjournals.annualreviews.orgby University of Texas - Austin on 10/23/07. For personal use only.INTRODUCTIONTo a first approximation, research in human vision can be divided into three ar-eas of investigation. Low-level or early vision is concerned with extraction ofphysical properties such as depth, color, and texture from an image as well asthe generation of representations of surfaces and edges (Marr 1982).Intermediate-level vision concerns extraction of shape and spatial relationsthat can be determined without regard to meaning but that typically require aselective or serial process (Ullman 1996). Finally, high-level vision concernsthe mapping from visual representations to meaning and includes the study ofprocesses and representations related to the interaction of cognition and per-ception, including the active acquisition of information, short-term memoryfor visual information, and the identification of objects and scenes. In thischapter we review three important areas of investigation in the study of high-level scene perception. First, we examine eye movements in scene perception,focusing on the cognitive control of eye movements and the degree to whichmeaning and ongoing cognitive processes influence eye movement behavior.Second, we review recent work on the nature of the scene representation that isretained across a saccade and other similarly brief intervals during ongoingscene perception. Finally, we review work on the interaction of cognition andperception, focusing on object and scene identification. Although these topicshave a long tradition of empirical investigation, they each have received aflurry of new work in the past few years.In research on high-level scene perception, the concept of scene is typicallydefined (though often implicitly) as a semantically coherent (and often name-able) view of a real-world environment comprising background elements andmultiple discrete objects arranged in a spatially licensed manner. Backgroundelements are taken to be larger-scale, immovable surfaces and structures, suchas ground, walls, floors, and mountains, whereas objects are smaller-scale dis-crete entities that are manipulable (e.g. can be moved) within the scene.Clearly, these definitions are neither exact nor mutually exclusive. For exam-ple, the distinction between a scene and an object depends on spatial scale. Anoffice scene may contain a desk as one of its component objects. But in a morefocused view, the desktop might become a scene, with its surface forming thebackground and a stapler, phone, and pen serving as individuated objects. It isdifficult to determine precisely when the spatial scale becomes too small or toolarge to call the resulting view a scene. Is the inside of a desk drawer a scene? Isa box of paperclips a scene? Most research on scene perception has avoidedthis problem of definition by using views of environments scaled to a humansize. So an encompassing view of a kitchen or a playground would be consid-ered a good scene, whereas a view of a box of paperclips or an aerial view of acity would not. For the current purposes we adopt this imprecise, intuitive, and244 HENDERSON & HOLLINGWORTHAnnu. Rev. Psychol. 1999.50:243-271. Downloaded from arjournals.annualreviews.orgby University of Texas - Austin on 10/23/07. For personal use only.not wholly satisfying definition, holding to the belief that definitions are oftenbest refined as a product of empirical investigation.EYE MOVEMENT CONTROL IN SCENE PERCEPTIONBecause of the optical structure of the eyes, the gradient in cone density in theretina, and the preferential mapping of foveal photoreceptors onto visual corti-cal tissue, acuity is highest at the point of fixation and drops off precipitouslyand continuously with increasing visual eccentricity (Anstis 1974, Riggs1965). The highest-quality visual information is acquired from the region ofthe scene that projects to the fovea, a region of the retina corresponding toabout the central 2º of the viewed scene (about the size of a thumbnail at arm’slength). The human visual-cognitive system takes advantage of the high re-solving power of the fovea by reorienting the fixation point around the viewedscene an average of three times each second via
View Full Document
Unlocking...