The Science of Color in Nature: a Data-Driven Approach to Color Harmony

This master's thesis investigates the relationship between image appreciation and statistical properties, extending previous research on grayscale images to color images [1]. The study analyzes natural color distributions using coarse-graining, color quantization, and a density-based approach. It also tests Moon and Spencer’s color harmony theory [2] in both natural and human preferences based contexts. Results reveal a bimodal Gaussian distribution of hues in nature and show that only the Contrast region of Moon and Spencer’s theory is consistently populated. However, survey data indicate that harmony perception depends on the reference hue, challenging the theory’s universality. This motivates the development of a more general and quantitative model of color appreciation. Additionally, the study introduces a new hue spectrum that better represents natural colors. This has applications in multiscale relevance, allowing for improved image representativeness assessments and insights into image processing and physical interpretability of machine learning dynamics. [1]Samy Lakhal, Alexandre Darmon, Jean-Philippe Bouchaud, and Michael Benzaquen. “Beauty and structural complexity”. In: Physical Review Research 2.2 (June 2020). issn: 2643-1564. [2] Parry Moon and Domina Eberle Spencer. “Aesthetic measure applied to color harmony”. In: JOSA 34.4 (1944), pp. 234–242.