Complex Object And Scene Perception

The goal of the research was to understand the role of contextual constraint in the visual analysis of complex, natural scenes. This issue bears directly on the basic architecture of the human visual system and has implications for the design of artificial vision systems devoted to object and scene analysis. Two hypotheses were considered: (1) Scene constraint influences the perceptual identification of individual objects; and (2) Scene constraint influences only post-identification object analysis. The main results of the research support the first hypothesis: Scene constraint does not directly influence perceptual analysis of component objects in human vision. Other results from the research supports the conclusion that semantic information is not used to drive initial eye movements in a scene, but does influence initial fixation time in a region and region refixation probability. Based on these results. a model of scene analysis was developed in which object identification is functionally isolated from scene meaning and gaze control is initially independent of scene semantics but becomes sensitive to meaning as scene perception unfolds over time. Continuing work is currently aimed at testing this model in human vision and gaze control, and implementing an artificial gaze control system on a working robot platform using a Markov Decision Process framework.