Regarding Scenes

When we view the visual world, our eyes flit from one location to another about three times each second. These frequent changes in gaze direction result from very fast saccadic eye movements. Useful visual information is acquired only during fixations, periods of relative gaze stability. Gaze control is defined as the process of directing fixation through a scene in real time in the service of ongoing perceptual, cognitive, and behavioral activity. This article discusses current approaches and new empirical findings that are allowing investigators to unravel how human gaze control operates during active real-world scene perception. KEYWORDS—scene perception; real-world scene; visual saliency; eye movements; gaze control; visual context It has been known for at least 30 years that the gist of a scene can be apprehended very rapidly, well within the duration of a single fixation or period of relative gaze stability (Potter, 1976). It has been known for even longer that viewers tend to move their eyes through scenes when they look at them (Buswell, 1935; Yarbus, 1967; see Fig. 1). Given fast gist understanding, why do we bother to move our eyes? Recent studies of change detection, object identification, and scene memory show that close or direct fixation is necessary to perceive local visual details, to unambiguously identify objects, and to encode object and scene information into shortand long-term memory (Henderson & Hollingworth, 1999; Hollingworth & Henderson, 2002). What we see and understand about the visual world is tightly tied to where our eyes are pointed. Why is fixation critical to these perceptual and cognitive processes? First, high-resolution visual information is acquired from only a very limited region of the scene surrounding the fixation point, with visual quality falling off precipitously and continuously from central vision into a low-resolution visual surround. The high resolving power of central vision is partly a consequence of the optical and anatomical structure of the eye and retina. Also, ‘‘cortical magnification’’ preferentially maps central vision onto the visual cortex, ensuring that more computational power is devoted to fixated regions. Therefore, to acquire high-quality visual input from a scene region, fixation must be directed to it. Second, there is a very tight link between attention and fixation. Although visual-spatial attention can be dissociated from where the eyes are fixated in laboratory demonstrations, attention is typically directed to the fixated location and the location to be fixated next. Attention is time-locked to eye-movement dynamics as eye-movement control circuits hold fixation and then release the eyes to the next fixation site. This time-locked relationship between shifts in attention and gaze appears to be mandatory, in part due to the tight neural integration of systems that control covert attention and those that control eye movements. Given the importance of fixation for perceptual and cognitive processing during scene perception, a critical issue concerns the representations and processes that govern where and for how long the eyes are directed to a particular scene region. This issue has become the focus of intense investigation in the past few years, with recent research emphasizing two general classes of factors that may drive gaze: bottom-up image properties, and cognitive knowledge structures used in a top-down manner.