Abstract
Recent findings in twin children suggest a heritable component for fixation behaviour towards social1 and complex2 scenes, revealing systematic variability in gaze behaviour. However, the nature of these differences and their persistence into adulthood are largely unclear. Here, we present results of an experiment tracking the gaze of 52 adults freely viewing 700 complex scenes for 3s each. We harnessed semantic annotations for >5,500 objects3 to probe the magnitude and consistency of individual differences in semantic salience3,4. Observers showed large and surprisingly consistent differences in the average number of objects they fixated and in their average dwell time per object (split half reliability for both r=.99). We further investigated the proportion of fixations towards objects from different semantic categories. This revealed considerable and highly consistent (r=.73-.98) individual differences in semantic salience for eight out of twelve tested categories: faces, emotional expressions, food, text, watchable objects, implied motion, smell, touch. For faces, this extended to an analysis limiting the data to the first fixation landing on each image (r=.79). Furthermore, the spatial distribution of faces drawing first fixations consistently varied between observers (r=.94 for average eccentricity). Our findings suggest that gaze behaviour is marked by attentional 'fingerprints': highly individual tendencies for visual exploration and for fixations along dimensions of semantic saliency. At least for faces this seems to extend to 'salience' in a stricter sense, suggesting the existence of a face priority map with individually different extent. This opens a new window for the study of attentional selection and suggests a potential link between individual differences in perception, personality and behaviour. 1. Constantino et al. Nature 547, 340–344 (2017). 2. Kennedy et al. Curr. Biol. 27, 3554–3560.e3 (2017). 3. Xu et al. J. Vis. 14, (2014). 4. Henderson & Hayes Nat. Hum. Behav. 1, 743–747 (2017).
Meeting abstract presented at VSS 2018