Infant egocentric scene statistics are simpler than adult egocentric and canonical natural scenes

Poster Presentation 53.473: Tuesday, May 19, 2026, 8:30 am – 12:30 pm, Pavilion
Session: Development

Philip McAdams1, Alexis Colwell1, Minju Kim1, Linda B Smith1; 1Indiana University

Vision emerges in a richly structured world. Classic work on natural scene statistics has quantified framed outdoor photographs, revealing scale-invariant amplitude spectra and characteristic local phase and contour structure, orientation anisotropies, and distinctive distributions of luminance and chromatic contrast. These statistics, part of our evolutionary experience, are thought to constrain early visual representations and approximate the input that shapes these representations during development. Yet infants and adults in industrialised societies spend ~90% of waking time indoors, and laboratory studies sample only a tiny fraction of this visual diet and rarely include naturalistic egocentric views. Building on recent work, we ask whether the everyday egocentric visual experiences of infants and adults share the same statistics as canonical natural scenes. We use head-mounted cameras to quantify the visual diet of 96 infants (1–24 months) and 20 adults in the US. We analyse ~100,000 primarily indoor (95%) egocentric images and compare them with two outdoor natural-scene corpora: adult egocentric views and natural-scene databases (e.g., McGill). We standardised images and computed image statistics linked to early visual processing (edge density/orientations, spectral slope, fractal dimension, luminance and colour), analysed with linear mixed-effects models and PCA-based comparisons across corpora. Infant and adult everyday egocentric visual experiences are simpler than egocentric outdoor and natural-scene databases. Simplicity is most pronounced in the first year, with scenes showing low edge density, edge-orientation entropy and fractal dimension, steeper spectral slopes (β=-1.6), and constrained luminance and chromatic distributions. Complexity increases in 12–24-month-olds, who encountered the brightest scenes and maximal luminance/chromatic contrasts relative to infant or adult egocentric groups. Thus, the “visual curriculum” shaping early visual cortex departs from canonical natural-scene statistics: it starts simpler, becomes more adult-like by the second year, yet remains simpler than natural scenes. Theories of visual perception and development must reconcile evolutionary constraints with egocentric daily-life inputs.