Low-level Visual Statistics in Infant-Perspective Scenes Change with Development

AbstractNumerous theories in statistical learning assume a data set for learning and learner that are more-or-less constant. The visual systems of human infants and their motor capabilities undergo dramatic postnatal development however. Here we used a large corpus of head camera images (50,706) collected by infants at 1-3 and 6-8 months of age to ask whether the visual scenes experienced by infants of different ages show bias towards information capable of being processed by the immature visual system. Using human coders to identify sparse scenes that could reveal whether younger infants are orienting to more visible content, we determined that the 1- to 3-month-old infants’ images included more visible information. This shows, for the first time, that very young infants visual preferences – demonstrated in the laboratory – directly influence the kinds of scenes they experience in everyday life. The findings also show that low-level natural scene statistics, rather than being universal, may change with development.

Return to previous page