Aesthetic Image Statistics Vary with Artistic Genre

Area of Science:

• Computer Vision
• Art Analysis
• Psychophysics

Background:

• Previous research has not established a universal link between low-level image statistics and general aesthetic ratings.
• Understanding visual features that influence aesthetic judgments in art remains an ongoing challenge.

Purpose of the Study:

• To investigate if different image statistics are crucial for artistic images across various genres.
• To identify image statistics that reliably correlate with aesthetic ratings using partial least squares regression (PLSR).

Main Methods:

• Estimated Fourier spectral slope, fractal dimension, and Shannon entropy for paintings categorized by subject matter (landscapes, people, still life, portraits, nudes, animals, buildings, abstracts).
• Conducted separate analyses on luminance and color information within the images.
• Applied PLSR to correlate image statistics with aesthetic ratings.

Main Results:

• PLSR models explained up to 75% of the shared variance between image statistics and aesthetic ratings.
• The most significant image statistics and image planes (luminance vs. color) varied considerably across different art genres.
• Identified genre-specific correlations between visual features and aesthetic appeal.

Conclusions:

• Image statistics are genre-dependent predictors of aesthetic ratings in art.
• Variations in visual statistics may correspond to specialized neural processing for different image types.
• This research provides insights into the quantitative analysis of aesthetic properties in art.