Grounding Color: Links Between Image-Based Color Features and Sensory Ratings

Yufeng Wu, Meichun Liu

Abstract: This study examines whether objective, imagederived color statistics align with human sensory ratings for color phrases drawn from six basic hues (black, white, red, green, blue, yellow), and whether these alignments persist beyond general concreteness. For each phrase, low-level visual features were extracted from associated images and linked to normed sensory ratings (visual, haptic, gustatory, olfactory, auditory, interoceptive; plus aggregate indices). Pairwise Spearman correlations with FDR control revealed a coherent pattern across colors: stable chromatic structure, especially the blue-yellow opponent centroid and saturation, was positively associated with visual and aggregate perceptual strength, whereas luminance variability was negatively associated. Selective positive relations between color entropy and gustatory/olfactory ratings were also observed. Concreteness was modeled as a continuous covariate via partial Spearman estimates; shifts were predominantly modest and attenuating, indicating that concreteness explains some shared variance without eliminating core color-sensory links. Exploratory moderation showed limited reliable effects after FDR control. Findings support an embodied account in which interpretable, low-level visual regularities provide compact, reproducible signals of modality-specific grounding across multiple color families, suitable for intelligent data-mining pipelines and multimodal NLP.

External IDs:dblp:conf/bigdataconf/0002L25