Guidance of Visual Attention by Low- and High-Level Features in Real-World Scenes

Marc Pomplun
University of Massachusetts Boston


Abstract: Several studies have shown that low-level visual features in naturalistic scenes, such as color or contrast, guide attention during inspection and search. I will present the results of our computational modeling studies showing that this guidance depends on the composition of a scene across feature dimensions. Furthermore, high-level features such as object and scene semantics are known to influence the allocation of attention as well. I will discuss our recent studies showing that even the semantic relations among scene objects are significant predictors of shifts of attention. Integrating these guidance mechanisms operating at various levels into one model may lead to a more comprehensive understanding of attentional control in real-world scenes.

Conference Sponsors