Abstract
Introduction: The extent to which eye movements are influenced by low-level saliency vs. other processes during free viewing of scenes (in the absence of any task) is debated (Koehler, 2014). Recent studies (Henderson et al., 2017, Peacock et al., 2019) suggest that eye movements during free viewing are directed to regions rated as semantically meaningful. Here, we compare eye movements during free viewing to those during a scene description, object search, and salient object search tasks. Methods: Two hundred participants were equally split across four conditions (free viewing, scene description, free viewing, object search, and saliency search) while viewing the same forty images. Images were presented for 2 seconds while eye position was measured. Objects within each image were categorized by independent raters on their relevance to the understanding of the scene and their saliency. Results: The mean correlation of fixation maps between free-viewing and scene description (r=0.67) was significantly greater (p<0.0005) than the correlation between free viewing and object search fixation maps (r=0.32) and greater (p<0.0005) than the correlation between free viewing and saliency search fixation maps (r=0.51). Participants fixated more frequently on objects that are relevant to scene understanding than other objects in the free viewing (fixation number difference=0.25, p<0.01) and scene description (fixation number difference=0.35, p=0.01). The target object was fixated on more frequently in the object search. There was a trend to fixate the most salient object over other objects in the saliency search but it did not reach significance (p=0.16). Conclusions: Our findings suggest that eye movement exploration during free viewing is most similar to a scene description task with fixations falling on regions that are important to understanding the scene.