Abstract
Background: It was shown that when observers were asked to rotate the color gamut of images of unfamiliar paintings to select the preferred one, they preferred the color compositions very close to the original paintings (Nascimento et al., 2017; Kondo et al., 2017). This study aims to explore what features of the color compositions underlie such original-preferred judgement for art paintings. Method: 4-AFC paradigm was used to measure the preference for art paintings. Observers (N=52) were asked to select the most preferable one among four images: original (0 deg) and three hue-rotated images (90, 180 and 270 deg) which had the same luminance and mean chromaticity as the original. In addition to the original condition C1, we tested spatial scrambling condition C2 (the images were divided into small square parts and scrambled), mixture condition C3 (the square parts were randomly selected from 20 different images) and hue-randomized condition C4 (the square parts were randomly rotated in hue angle). Results and Discussions: Originals were selected most frequently in C1-3 conditions, suggesting no or little contextual effect on preference. Furthermore, original-preferred judgement even in the mixture condition C3 implies a certain common color structure among art paintings. We therefore analyzed the color statistics in CIELAB of 5,591 art paintings of various categories and found that paintings share the common features regardless of the categories, that is, positive skewness of red-green (a*) and positive correlation between lightness (L*) and blue-yellow (b*). Corresponding to this, regression analysis between preference and color statistics revealed that selection probability of the images measured in the 4-AFC experiment significantly correlated with skewness of a* and correlation between L* and b* of the images. These findings suggest that implicit criteria underlie both in creation and aesthetic judgment of art paintings, probably relating to naturalness and/or novelty of color composition.
Meeting abstract presented at VSS 2018