Abstract
Most of the visual field is peripheral, and, compared to the fovea, the periphery encodes visual input with less fidelity. What information is encoded and what is lost in the visual periphery? A systematic way to answer this question is to determine how sensitive the visual system is to different kinds of lossy image perturbations. A difficulty of this approach is that in addition to
information loss in the visual system, there are other factors that can reduce performance in behavioural experiments; for example, task performance may be limited by cognitive factors such as attention or memory. Here, we develop and explore an experimental paradigm that
probes the detectability of perturbations of natural image structure with high sensitivity. Observers compared modified images to original natural scenes in a temporal three-interval oddity task. We consider several lossy image transformations, including Gaussian blur and textures synthesised from the Portilla and Simoncelli algorithm. While our paradigm demonstrates metamerism (physically different images that appear the same) under some
conditions, in general we find that contrary to an extreme "lossy representation" account of peripheral encoding, humans can be capable of impressive sensitivity to deviations from natural appearance. The results force us to consider richer representations of peripheral image
structure.
