The Predictive Ability of GBVS Feature Channels on Infants' Fixations of Natural Scenes

For decades, there has been significant interest in the role of salience on infants' looking behavior. Previous work has shown that computational models of salience can predict where infants will look within a given scene, but these models combine information across multiple different feature dimensions. It is thus unclear how specific visual stimulus dimensions contribute to infants' looking. To address this gap in knowledge, we measured the locations fixated by 4- to 12-month-old infants (N = 167) while viewing photographs of natural scenes, and we evaluated the predictive ability of the orientation, color, and intensity channels within the Graph-Based Visual Salience model (GBVS). We found that infants' fixations were primarily predicted by orientation and that this effect strengthened with age. However, we also found that the effects of orientation and color varied across different stimulus sets, with exploratory analyses suggesting that these differences are driven by differing levels of visual clutter.