Neural dissociation between computational and subjective image complexity

The complexity of images plays an important role in visual perception, including aesthetic appreciation. We here propose the information-based model of visual complexity (IVC) based on multiscale entropy and nonlossy run-length compression. We show that the IVC model accounts for a large amount of variance in people’s complexity ratings of natural scenes. To determine if the IVC model captures the same kind of information that people use in their ratings, we analyzed the public BOLD5000 set of functional magnetic resonance image scans of four participants viewing almost 5,000 images. We constructed regressors from the IVC predictions for these images and the complexity ratings averaged over 50 participants. We find that a large proportion of voxels in early visual cortex significantly relates to predictions of the IVC model, but only few voxels relate to complexity ratings. In high-level visual brain regions, we see the opposite pattern—we find many voxels relate to complexity ratings but only few relate to the IVC model. The relationship between complexity and aesthetic liking further illuminates this dissociation: We find that aesthetic liking ratings are strongly correlated with complexity ratings, but not IVC model predictions. We conclude that the relatively low-level IVC model, although correlating well with people’s complexity ratings, does not fully capture the range of information that people use in judging image complexity. Higher level information that is represented beyond early visual cortex plays an important role in perceived image complexity. (PsycInfo Database Record (c) 2023 APA, all rights reserved) (Source: journal abstract)