Academic journal article Perception and Psychophysics

Integration of Multiple Views of Scenes

Academic journal article Perception and Psychophysics

Integration of Multiple Views of Scenes

Article excerpt

In two experiments, memory was tested for changes in viewpoints in naturalistic scenes. In the key study condition, participants viewed two images of the same scene from viewpoints 40° apart. There were two other study conditions: The two study images were identical or were of different scenes. A test image followed immediately, and participants judged whether it was identical to either of the study images. The scene in the test image was always the same as in a study image and was at least 20° from any study image on different trials. Two models were tested: (1) views stored and retrieved independently and (2) views combined at retrieval. The crucial test of these hypotheses involved a comparison (in the key study condition) of the interpolation condition (the test image was presented between the two study images and 20° from both) and the extrapolation condition (it was 20° from one study image and 60° from the other). Performance in the interpolation condition was far worse than what was predicted by the first model, whereas the second model fit the data quite well. The latter model is parsimonious in that it integrates previous experiences without requiring the integration of the views in memory. We review some of this model's broader implications.

One of the most amazing properties of the visual system is its ability to identify our surroundings, despite the variety of visual conditions that change the image reflected onto the retina (e.g., lighting, color). This is especially apparent in our ability to recognize the same scenes from a number of different views. Although there have been many studies on the recognition and perception of scenes over the past few decades (for a review, see Henderson, 2007), only a few have investigated how scenes are represented across different viewpoints and how the system uses this information in scene recognition (Christou & Bülthoff, 1999; Garsoffky, Schwan, & Hesse, 2002; Hock & Schmelzkopf, 1980; Nakatani, Pollatsek, & Johnson, 2002; Shelton & McNamara, 1997). Christou and Bülthoff focused on this problem by using a navigation task in a virtual-reality setting. Participants were allowed to explore an attic (consisting of multiple rooms) from certain viewpoints. When participants were asked to recognize still images taken from this environment, it was shown that scene recognition was highly viewpoint dependent. However, the scenes were relatively impoverished, because they consisted mainly of rooms defined by planar walls, floors, and ceilings, with only the angles and orientations of these planes able to provide information for distinguishing these views. Shelton and McNamara were also interested in the representation of large, navigable spaces and had participants memorize the relative locations of items from specific viewpoints within an array of objects. Participants were then asked to imagine themselves aligned to different views and to report the relative positions of the objects. Imagined headings aligned with the study views were easier to retrieve than novel headings, and so performance was consistent with a viewpoint-dependent representation of the space. The important thing to note about these two studies is that the space, as represented by the walls or the array of objects, was the primary focus of the investigation.

In other studies (Garsoffky, Huff, & Schwan, 2007; Garsoffky et al., 2002; Huff, Schwan, & Garsoffky, 2007), participants studied dynamic scenes depicting moving players in a soccer game (Garsoffky et al., 2002) or a basketball game (Garsoffky et al., 2007). In these studies as well, it was found that performance was viewpoint dependent. However, as discussed in Huff et al. (2007), the nature of representing dynamic scenes may be likened more to a descriptive "event" representation (Zacks, Tversky, & Iyer, 2001) and so may involve consideration of information not present in the static scenes that are typical of scenerecognition studies. …

Search by... Author
Show... All Results Primary Sources Peer-reviewed

Oops!

An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.