Anna Shafer-Skelton¹, Mary Hayhoe¹; ¹University of Texas at Austin

To understand the visual system, it is critical to understand its inputs. Early visual regions are sensitive to orientation and spatial frequency information (crisp and blurry oriented edges), and classic work has quantified this information with Fourier spectra in both hand-selected natural-scene photographs and image sets structured by scene category. While some work has analyzed the information present at fixation during several naturalistic tasks, the release of large data sets has only recently made it possible to capture gaze-centered image statistics across the wider visual field. Here, we analyzed stills taken from the Visual Experience Dataset (VEDB; Greene et al., 2024), which recorded eye-tracking data alongside wide-angle egocentric video as participants engaged in a broad range of real-world activities. We captured 8,781 stills and analyzed the gaze-centered vs. head-centered Fourier spectra of each, across both a smaller and larger field of view. We summarized Fourier information using bins of four orientations and two spatial frequencies. Initial analyses show stronger cardinal information in head-centered vs. gaze-centered stills, especially at lower spatial frequencies and for the larger field of view. Ongoing analyses investigate how tasks, settings, and choice of image statistics influence gaze-centered vs. head-centered differences. This work demonstrates an important step towards understanding how our real-world visual inputs interact with the retinotopic (gaze-centered) visual system, as well as the impact of active sampling behavior on these inputs.

Acknowledgements: This work was supported by an NEI postdoctoral NRSA fellowship to AS (5F32EY036266-02)