UNDERSTANDING AND PREDICTING HUMAN VISUAL ATTENTION

Abstract

An understanding of how the human visual system works is essential for many applications in computer vision, computer graphics, computational photography, psychology, sociology, and human-computer-interaction.

To provide the research community with access to easier, cheaper eye tracking data for developing and evaluating computational models for human visual attention, this thesis introduces a webcam-based gaze tracking system that supports large-scale, crowdsourced eye tracking deployed on a crowd-sourcing platform. By using this tool, we also provide a benchmark data set to quantitatively compare existing and future models for saliency prediction. To explore where people look while performing complicated tasks in an interactive environment, we introduce a method to synthesize user interface layouts, present a computational model to predict users' spatio-temporal visual attention for graphical user interfaces, and show that our model outperforms existing methods. In addition, we explore how visual stimuli affect brain signals extracted by fMRI.

Our tool for crowd-sourced eye tracking, a large data set for scene image saliency, models for user interface layouts synthesis and visual attention prediction and study for visual stimuli driven change of brain connectivity should be useful resources for future researchers to create more powerful computational models for human visual attention.