Ece Yatikci¹, Ben Lonnqvist¹, Clementine Levy-Fidel¹, Marco Bertamini², Michael Herzog¹; ¹Laboratory of Psychophysics, Brain Mind Institute, École Polytechnique Fédérale de Lausanne (EPFL), Lausanne, Switzerland, ²Department of General Psychology, University of Padova, Padova, Italy

Peripheral vision is often assumed to be a low-resolution version of the fovea. While blur clearly contributes to the loss of fine detail in the periphery, it is unclear whether other mechanisms are also involved in peripheral perception. The Honeycomb illusion illustrates this loss, as small lines (barbs) that are visible in the fovea disappear in the periphery, while the underlying grid texture remains visible (Bertamini et al., 2016). This disappearance is often attributed to low-pass filtering. To test the role of blur, we tested performance with the Honeycomb illusion and with a visual pop-out search task in human observers (N = 20). We fitted parameters of a blur model to the Honeycomb illusion data and evaluated the same model on the search task. While blur captured key aspects of barb disappearance in the Honeycomb illusion, it failed to account for human performance in visual pop-out search. To test whether representational compression could explain peripheral perception, we trained autoencoders with latent bottlenecks, reflecting reduced cortical resources in the periphery. Two of the four models reproduced barb disappearance in the Honeycomb illusion, but all failed on pop-out search as the odd man out target was not preserved in the latent representation. Our findings show that no single model is sufficient to account for peripheral perception. Phenomena that appear compatible with simple low-pass filtering, as in the Honeycomb illusion, may depend on additional mechanisms, including compression.