Textures are less comfortable to view than other natural images, consistent with cortical hyper-metabolism theory

Abstract

Images considered uncomfortable to view often over-represent mid-range spatial frequencies (mid-SFs) relative to the average 1/f amplitude spectrum of natural images. A proposed mechanism for this discomfort known as the Cortical Hyper-metabolism Hypothesis (CHH) posits that such images over-activate the visual cortex, which is highly sensitive to mid-SFs. We tested a novel prediction of the CHH, namely that natural image categories like textures which over-represent mid-SFs will be less comfortable to view on average than other natural images. We indeed found that texture images were less comfortable to view than various non-textures, validating this prediction of the CHH and demonstrating its potential relevance to natural image aesthetics. Passing our images through models of early visual neurons demonstrated that the mean activity of a population of simulated log-Gabor filters resembling V1 simple cells best predicted discomfort. Manipulations which selectively altered the image phase or amplitude spectra while leaving the other intact were less comfortable than the originals, suggesting that visual comfort is not a simple function of either one of these characteristics. Our results suggest that although many factors contribute to the visual comfort of natural image categories, early visual processing may play an important role, consistent with the CHH.