Direct perception of affective valence from vision

Area of Science:

• Cognitive Neuroscience
• Computer Vision
• Psychology

Background:

• Subjective feelings (valence) are traditionally linked to conceptual and bodily states.
• The role of objective visual environmental statistics in shaping emotional valence is less understood.

Purpose of the Study:

• To investigate if the valence of feelings can be decoded directly from objective ecological statistics of the visual environment.
• To explore the contribution of visual valence (VV) to subjective emotional experience.

Main Methods:

• Trained a machine learning model (VV model) on low-level image statistics from emotionally charged photographs.
• Tested the VV model's predictive accuracy on human valence ratings and abstract paintings.
• Examined human observers' valence experience with and without conceptual image analysis.
• Investigated the neural basis of VV in visual brain regions using deep generative networks.

Main Results:

• The VV model accurately predicted human valence ratings and generalized robustly to abstract art.
• Limiting conceptual analysis in humans increased the influence of VV on valence experience, aligning with machine perception.
• Neural activity in lower to mid-level visual areas correlated with VV.
• Synthesized images in generative networks reflected positive versus negative VV.

Conclusions:

• Subjective valence experience has at least two distinct modes: one indirect (meaning-based) and one direct (ecological statistics-based).
• Ecological statistics in the visual environment can be directly perceived as an objective property contributing to subjective valence.
• This suggests a direct perception pathway for emotional valence embedded in visual input.