Aesthetic preference for art can be predicted from a mixture of low- and high-level visual features.
Level 4 - case-series / case-control
Level by design analogy (cross-sectional behavioral testing and computational modeling), not clinical CEBM.
PubMed 34017097 · doi:10.1038/s41562-021-01124-6
What was done
Researchers developed and evaluated a computational framework to test whether aesthetic value ratings for visual art can be predicted from image features. They used regression models containing a combination of low- and high-level visual features to model subjective preference ratings within and across individuals, and assessed whether these predictive features emerge hierarchically in a deep convolutional neural network trained on object recognition.
What was found
The abstract reports no numerical data, effect sizes, sample sizes, or model performance metrics. It qualitatively reports that subjective value ratings for art were successfully predicted within and across individuals using shared interpretable features, and that these predictive features arose hierarchically in an object-recognition neural network.
Why it matters
This study provides evidence that aesthetic appreciation of art is not entirely arbitrary, but can be systematically linked to hierarchical visual properties of images.
Limits
The abstract provides no sample sizes, participant demographics, art genre details, or quantitative prediction accuracies. External validity across diverse cultural contexts and non-visual influences on aesthetic value were not assessed in the abstract.
Cited by
- supports Computational neural network models decomposing low-level visual features of abstract art, such as colors and visual composition, can decode or predict whether human observers find the art pleasant or liked.