Aesthetic preference for art can be predicted from a mixture of low- and high-level visual features

Abstract

It is an open question whether preferences for visual art can be lawfully predicted from the basic constituent elements of a visual image. Here, we developed and tested a computational framework to investigate how aesthetic values are formed. We show that it is possible to explain human preferences for a visual art piece based on a mixture of low-and high-level features of the image. Subjective value ratings could be predicted not only within but also across individuals, with a regression model with the same set of interpretable features. We also show that the features which predict aesthetic preference can emerge hierarchically within a deep convolutional neural network trained only for object recognition. Our findings suggest that human preferences for art can be explained at least in part as a systematic integration over the underlying visual features of an image.

Functional specificity and neural integration in the aesthetic appreciation of artworks with implied motion

“…Neuroaesthetics research has focussed on deconstructing features of visual art, such as form, colour, symmetry, complexity, luminance and contrast (Bar & Neta, 2006;Bona et al, 2015;Graham et al, 2016;Hayn-Leichsenring et al, 2020;Iigaya et al, 2021;Jacobsen & Höfel, 2003;Jacobsen et al, 2006;Nadal et al, 2010;Palmer & Schloss, 2010;Van Geert & Wagemans, 2019;Vartanian et al, 2013). Although many different visual features of art have been studied from a neuroscientific perspective, it is maybe surprising that implied motion has only received limited attention (Di Dio et al, 2016;Kim & Blake, 2007;Thakral et al, 2012).…”

“…Given the complexity of aesthetic judgements, there is a need, therefore, to move into multivariate space (Kriegeskorte, 2009;Norman et al, 2006). Some neuroaesthetics work has started to approach questions using multivariate classifiers (Iigaya et al, 2021). More generally, human neuroscience research on functional integration or how multiple brain systems interact with one another for a given process (Bullmore & Sporns, 2009;Park & Friston, 2013) is particularly relevant here because we want to estimate how and when signals from distinct circuits are integrated in the process of constructing an aesthetic judgement.…”

Self-relevance predicts the aesthetic appeal of real and synthetic artworks generated via neural style transfer

“…Thus, self-relevance is a key determinant of aesthetic appeal, independent of artistic skill and image features.Recently it has been shown that a linear combination of low-and high-level image features predicts, on average, around 20% of variance in observer ratings [2]. A comparable performance (16.4%) was obtained by a deep neural network trained to predict aesthetic ratings [2]. These findings suggest that a large fraction of variance in aesthetic ratings across individuals remains unexplained.…”

“…Self-relevant, synthetic artworks were rated as more aesthetically appealing than matched control images, at a level similar to real artworks. Thus, self-relevance is a key determinant of aesthetic appeal, independent of artistic skill and image features.Recently it has been shown that a linear combination of low-and high-level image features predicts, on average, around 20% of variance in observer ratings [2]. A comparable performance (16.4%) was obtained by a deep neural network trained to predict aesthetic ratings [2].…”

“…Recently it has been shown that a linear combination of low-and high-level image features predicts, on average, around 20% of variance in observer ratings [2]. A comparable performance (16.4%) was obtained by a deep neural network trained to predict aesthetic ratings [2].…”