Abstract
When viewing natural images displayed on computer screens, observers show a considerable tendency to fixate their eyes more towards the center of the scene than any other image region regardless of its visual content. This so-called "center bias" is a well-known phenomenon that has been observed and examined in numerous studies on human eye movement behavior. Consequently, computational prediction of eye movements attempts to model the center bias in order to facilitate better estimation of the spatial distribution of human fixations. That being said, the center bias is often regarded merely a global spatial occurrence across complete scan paths. However, this view hinders any relation this bias may have with the dynamic nature of viewing. In the present work, we seek to uncover such relations by analyzing time-oriented gaze-paths both spatially and temporally. Indeed, both types of structures emerge, for example, in the way the center bias persists regardless of the current fixation location of an observer, or how it endures deep into the scan path. Based on these observations, we propose a more elaborate model of the center bias, that captures its manifestation in both space and time.