The Relation Between Individual Fixation Biases Towards Faces and Inanimate Objects

Individuals differ in the way they look at faces. Some prefer to fixate closer to the eye region, others tend to look near the nose or mouth. Previous studies proved these differences consistent, generalizing from image viewing in the lab to real-life interactions and to enhance individual face recognition performance. So far, the underlying mechanisms are unclear. The prevailing hypothesis posits that these biases are domain-specific and linked to individually shifted face-templates in retinotopic coordinates. However, gaze biases could also reflect domain-general individual visual field geometry, which would be expected to generalize across different object categories including faces. To juxtapose these hypotheses, we showed 52 participants 700 images of natural scenes from a publicly available image data set in an eyetracking experiment (Eyelink 1000 Tower Mount). For every object-directed fixation, we determined the relative height within the respective object using existing pixel masks of inanimate objects. Additionally, we created 2055 hand-delineated pixel masks for every human face, eye and mouth in the scenes, allowing us to determine face fixation heights relative to the vertical distance between eyes and mouth. Including replication datasets, we analyzed more than 0.8 million fixations toward faces and inanimate objects across 206 observers. Split-half correlations replicated previous findings of consistent individual fixation biases in faces and show that these also exist for inanimate objects (both r > .9). Most importantly, individual fixation heights were highly correlated between faces and inanimate objects (r > .7). Individual biases and the relationship between them were reliable across sessions separated by several weeks and proved robust in several control analyses for object size and eccentricity, suggesting that domain-general field biases shape individual viewing behavior. We aim to collect additional data on individual biases in recognition performance to validate our findings independently of eyetracker accuracy and possible related confounds.