Isolating task-related, low-level, and high-level influences on interobserver congruency of eye movements in scene viewing

When different individuals view the same image, their eye movements are often similar. The degree of this similarity is called interobserver congruency (IOC). We investigated how IOC is determined by three factors that impact gaze control: the visual features of an image, its semantic content, and the task observers perform when viewing it. We first quantified the low-level physical features and semantic complexity of images using a well-established measure of visual clutter and a novel measure based on a distributional semantic model, respectively. Then, we selected an image set in which these two measures were uncorrelated and used it in an eye-tracking experiment, in which observers either freely viewed the images or memorised their content. We assessed IOC using four metrics: two capturing purely spatial IOC and two sensitive to spatiotemporal IOC, i.e., considering both the locations and order of fixations. The values of the two spatial metrics decreased with increasing clutter and remained insensitive to the remaining factors we investigated. The two spatiotemporal metrics were sensitive to both clutter and semantic complexity, but, interestingly, in different ways. Specifically, according to a metric that was insensitive to the presence of objects in images, IOC decreased with clutter and increased with semantic complexity, while the metric sensitive to objects showed the opposite pattern. Additionally, we found that the memorization task rendered oculomotor behaviour more exploratory, which had modulatory effects on spatiotemporal - but not spatial - IOC. Together, our findings reveal how factors known to shape gaze behaviour determine IOC.