Guiding Attention in Real-World Scenes

点击翻译按钮获取中文摘要

Project Summary Real-world scenes contain far more information than we can perceive at any given moment. Scene perception therefore requires attentional selection of relevant scene regions for prioritized processing. How are those aspects of the world that should receive priority selected? Although much past research has focused on how attention is guided by the visual properties of a scene, new evidence from meaning maps, developed in the previous funding period, established that the distribution of meaning across a scene plays a central and often dominant role in guiding attention. This surprising finding raises many important new questions about the nature of scene meaning and its specific role in attentional guidance. The overarching goal of this project is to understand in detail how the semantic features of a scene’s objects and functional spaces influence the guidance of visual attention in complex real-world scenes. The specific aims are: (1) To determine the role of object semantics in attentional guidance in scenes; (2) To determine the role of functional spaces in attentional guidance in scenes; (3) To determine how viewing task interacts with scene semantics in guiding attention. The project is innovative in expanding the traditional study of attention to explicitly consider the role of meaning. To this end, new semantic maps capitalizing on the meaning map concept will be used capture local region meaning continuously over a scene, allowing for direct investigation of the relationships of different types of meaning with attention. The project is innovative in (1) expanding the traditional study of visual attention to explicitly consider the role of semantics; (2) focusing on the semantics of both scene content (objects) and scene structure (space); (3) considering the role of meaning in attentional guidance in the context of viewing task; (4) integrating the use of a wide variety of cognitive science methods marshalled in the service of understanding the influence of meaning on visual attention in real-world scenes, including eyetracking, large-scale crowd-sourcing, computational image processing, computational semantic modeling, and deep convolutional neural networks. The project is significant in challenging current models of visual attention to account for the role of scene meaning. Because the proposed studies test competing models, the results will lead to the development of integrative theoretical frameworks that advance the field regardless of the outcome. While focused on basic science, the studies have potentially important translational implications by providing a more complete characterization of the processes associated with visual attention. The proposed studies may ultimately lead to the development of rehabilitation strategies for visual attention as it operates in the real world, better capitalizing on the use of a viewer’s knowledge to offset disrupted functions in those with attention and vision deficits.