Abstract
Models that aim at predicting fixation locations of observers in a natural scene can be based on assumptions on bottom up processes, top down processes, and systematic eye movement tendencies. Among the best predictors of fixation locations in an image is the distance of an image location from the image center. Because this central fixation bias (Tatler, 2007) is independent of image content, initial fixation position, screen or eye position, it is a very strong nuisance factor for scene viewing experiments under laboratory conditions. Scene preview from as short as 75 ms has been shown (Vo & Henderson, 2010) to influence saccade target selection effectively. Thus, a short scene preview might alter eye guidance during initial fixations and result in a reduced central fixation bias. In a scene viewing experiment, we manipulated the initial fixation position to keep the eyes for a certain amount of time fixated on one location before they were allowed to explore the scene freely for five seconds. As a result, we found that the central fixation bias was reduced for all pretrial fixation times from 125 ms to 1000 ms compared to a control group without scene preview. There were no systematic differences between the different pretrial fixation times. Our results have important practical implications for the evaluation of models of visual attention, since controlled initial fixation durations reduce attentional capture of the stimulus and, therefore, the central fixation bias.
Meeting abstract presented at VSS 2016