Visual features for scene recognition and reorientation

Abstract

In this thesis, I investigate how scenes are represented by the human visual system and how observers use visual information to reorient themselves within a space. Scenes, like objects, are three-dimensional spaces that are experienced through twodimensional views and must be recognized from many different angles. Just as people show a preference for canonical views of objects, which best show the object's surfaces and shape, people also show a preference for canonical views of scenes, which show as much of the surrounding scene layout as possible. Unlike objects, scenes are spaces which envelope the observer and thus a large portion of scene processing must take place in peripheral vision. People are able to perform many scene perception tasks, such as determining whether a scene contains an animal, quickly and easily in peripheral vision. This is somewhat surprising because many perceptual tasks with simpler stimuli, such as spotting a randomly-rotated T among randomly-rotated Ls, are not easily performed in the periphery and seem to require focal attention. However, a statistical summary model of peripheral vision, which assumes that the visual system sees a crowded, texture-like representation of the world in the periphery, predicts human performance on scene perception tasks, as well as predicting performance on peripheral tasks with letter stimuli. This peripheral visual representation of a scene may actually be critical for an observer to understand the spatial geometry of their environment. People's ability to reorient by the shape of an environment is impaired when they explore the space with central vision alone, but not when they explore the space with only peripheral vision. This result suggests that peripheral vision is well-designed for navigation: the representation in peripheral vision is compressed, but this compression preserves the scene layout information that is needed for understanding the three-dimensional geometry of a space.