Figure - available from: Nature Human Behaviour
This content is subject to copyright. Terms and conditions apply.
The LFS model successfully predicts the subjective value of paintings a, The LFS model with shared features captured in-lab participants’ art liking ratings. The predictive score, defined by the Pearson correlation coefficient between the model’s out-of-sample prediction and actual ratings, was significantly greater than chance for all participants who performed the task in the lab. The model was trained on six participants and tested on the remaining participant (blue), trained and tested on the same participant (red) and trained on on-line participants and tested on in-lab participants (yellow). In-lab participants performed a long task with 1,001 trials. Statistical significance was tested against a null distribution of correlation scores constructed by the same analyses with permuted image labels. The chance level (the mean of the null distribution) is indicated by the dotted lines (at 0). The same set of features (c) was used throughout the analysis. b, Our model also successfully accounted for the on-line participants’ liking of the art stimuli. We trained the model on all-but-one participants and tested on the remaining participant (left). We also fit the model separately to in-lab participants and tested it independently on all on-line participants (middle). The model predicted liking ratings significantly in all cases, even when we used low-level attributes alone (right). Each on-line participant performed approximately 60 trials. The error bars show the mean and s.e. over participants, while the dots indicate individual participants. The chance level (the mean of the null distribution constructed in the same manner as F) is indicated by the dotted line. c, Weights on shared features that were estimated for in-lab participants. We estimated weights by fitting individual participants separately. d, The low-level features can predict the variance of high-level features. Classification accuracy (high or low values, split by medians) is shown. Note that, although the prediction is highly significant, there is still a small amount of variance remaining that is unique to high-level features. The chance level (the mean of the null distribution) is indicated by the dotted line. The error bars indicate the standard errors over cross-validation partitions. In all panels, three asterisks indicates P < 0.001 against permutation tests.
Source publication
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-...
Similar publications
We process and integrate multiple timescales into one meaningful whole. Recent evidence suggests that the brain displays a complex multiscale temporal organization. Different regions exhibit different timescales as described by the concept of intrinsic neural timescales (INT); however, their function and neural mechanisms remains unclear. We review...
Citations
... These features can be categorized into lowlevel features and high-level features. Low-level features, such as luminance, colors, edges, and textures, are directly extracted from the external forms of the input image and are broadly used for assessing the aesthetics of paintings [7][8][9][10] and photographs [11][12][13][14][15]. High-level features representing semantic information are used to examine how semantic information affects the aesthetics of photographs [16,17] and paintings [18]. Attempts have been made to quantify the relationships between certain image features and personal aesthetic judgments of static generative art images [19,20]. ...
... To determine an appropriate sampling interval for noticeable perceptual differences between adjacent images within one sequence, a pre-experiment was conducted prior to the main experiment. Ten image sequences, that served as stimuli, were sampled from ten generative artworks, each with the same coloration, at intervals of 9,12,15,18,21,24,27,30,36, and 48 seconds, and contained ten static images. ...
Several automated aesthetic assessment models were developed to assist artists in producing artwork with high aesthetic appeal. However, most of them focused on static visual art, such as photographs and paintings, and evaluations of dynamic art received less attention. Dynamic visual art, especially computer-based art, includes diverse forms of artistic expression and can enhance an audience’s aesthetic experience. A model for evaluating dynamic visual art can provide valuable feedback and guidance for improving artistic skills and creativity, thereby benefiting audiences. In this study, we created eight generative artworks with dynamic art forms based on a commonly used method. We established a time-dependent model to predict the aesthetics based on visual features. We quantified the artworks according to selected image features and found that these features could effectively capture the characteristics of the changing visual forms during the generation process. To explore the effects of time-varying features on aesthetic appeal, we built a panel regression model and found that the aesthetic appeal of the generated artworks was significantly affected by their skewness of the luminance distribution, vertical symmetry, and mean hue value. Furthermore, our study demonstrated that the aesthetic appeal of dynamic generative artworks could be predicted by integrating image features into the temporal domain.
... Herein, we investigated the effect of congenital differences in colour vision on visual attention and viewers' impressions of complex images. Using artistic paintings, in which a range of visual features (from low to high levels) have been shown to explain impressions, we extracted a variety of impressions, including colour and aesthetic sensations [7][8][9][10]. We devised an experiment wherein trichromatic observers were shown images that simulated dichromatic perception (hereafter, 'simulated dichromats') to compare the experience of congenital dichromacy with the short-term experience of simulated dichromacy. ...
Humans exhibit colour vision variations due to genetic polymorphisms, with trichromacy being the most common, while some people are classified as dichromats. Whether genetic differences in colour vision affect the way of viewing complex images remains unknown. Here, we investigated how people with different colour vision focused their gaze on aesthetic paintings by eye-tracking while freely viewing digital rendering of paintings and assessed individual impressions through a decomposition analysis of adjective ratings for the images. Gaze-concentrated areas among trichromats were more highly correlated than those among dichromats. However, compared with the brief dichromatic experience with the simulated images, there was little effect of innate colour vision differences on impressions. These results indicate that chromatic information is instructive as a cue for guiding attention, whereas the impression of each person is generated according to their own sensory experience and normalized through one's own colour space.
... In the visual domain, we used landscapes as a reward and images of walls as an alternative ( Fig. 1a and Methods). Staring at landscapes has been rated positively in the past unlike looking at walls 7,17 . ...
... Although a large body of literature has been dedicated to the empirical study of primary and secondary rewards, the empirical study of intrinsic rewards is still in its infancy (for recent work see refs. 7,11,[81][82][83][84][85]. This line of research is challenging as intrinsic rewards are difficult to quantify, yet critical for a full and deep understanding of the human experience. ...
Humans frequently engage in intrinsically rewarding activities (for example, consuming art, reading). Despite such activities seeming diverse, we show that sensitivity to intrinsic rewards is domain general and associated with mental health. In this cross-sectional study, participants online ( N = 483) were presented with putative visual, cognitive and social intrinsic rewards as well as monetary rewards and neutral stimuli. All rewards elicited positive feelings (were ‘liked’), generated consummatory behaviour (were ‘wanted’) and increased the likelihood of the action leading to them (were ‘reinforcing’). Factor analysis revealed that ~40% of response variance across stimuli was explained by a general sensitivity to all rewards, but not to neutral stimuli. Affective aspects of mental health were associated with sensitivity to intrinsic, but not monetary, rewards. These results may help explain thriving and suffering: individuals with high reward sensitivity will engage in a variety of intrinsically rewarding activities, eventually finding those they excel at, whereas low sensitivity individuals will not.
... These techniques can help uncover a visual language for aesthetics. Given the universal mechanisms of human vision (Graham & Redies, 2010;Iigaya et al., 2021), humans may have evolved an ongoing readiness for aesthetic evaluations, which conferred evolutionary advantages, such as finding a potential mate or ideal habitat (Brielmann & Pelli, 2018;Chatterjee, 2022;Wassiliwizky & Menninghaus, 2021). ...
... Individual features have been extracted directly from photographic images, such as rectilinearity (Nasr et al., 2014) or the distribution of oriented spatial frequencies (Oliva & Torralba, 2001). In the context of aesthetics, Iigaya et al. relied on features extracted from deep neural networks (Iigaya et al., 2021). To our knowledge, our approach is the first that defines a comprehensive set of contour features in closed mathematical form on the basis of contours in complex, real-world scenes. ...
... Our results support the notion of a perceptual reward system (Biederman & Vessel, 2006;Vessel et al., 2021). Previous studies that have investigated the role of statistical image properties using global quantitative measures such as contrast, entropy, spatial frequency, and self-similarity (Graham & Redies, 2010;Iigaya et al., 2021;Palmer et al., 2013). In this study, we found that structural contour properties are remarkably predictive of aesthetic responses to natural scenes. ...
To what extent do aesthetic experiences arise from the human ability to perceive and extract meaning from visual features? Ordinary scenes, such as a beach sunset, can elicit a sense of beauty in most observers. Although it appears that aesthetic responses can be shared among humans, little is known about the cognitive mechanisms that underlie this phenomenon. We developed a contour model of aesthetics that assigns values to visual properties in scenes, allowing us to predict aesthetic responses in adults from around the world. Through a series of experiments, we manipulate contours to increase or decrease aesthetic value while preserving scene semantic identity. Contour manipulations directly shift subjective aesthetic judgments. This provides the first experimental evidence for a causal relationship between contour properties and aesthetic valuation. Our findings support the notion that visual regularities underlie the human capacity to derive pleasure from visual information.
... Generally, people's judgment on image aesthetics comes from the perception of various visual attributes [14]. As illustrated in Fig. 1, during image aesthetics rating, people first understand the image theme, based on which they proceed to analyze the visual attributes, e.g., interesting content, good lighting, vivid color and depth of field, etc. ...
People usually assess image aesthetics according to visual attributes, e.g., interesting content, good lighting and vivid color, etc. Further, the perception of visual attributes depends on the image theme. Therefore, the inherent relationship between visual attributes and image theme is crucial for image aesthetics assessment (IAA), which has not been comprehensively investigated. With this motivation, this paper presents a new IAA model based on Theme-Aware Visual Attribute Reasoning (TAVAR). The underlying idea is to simulate the process of human perception in image aesthetics by performing bilevel reasoning. Specifically, a visual attribute analysis network and a theme understanding network are first pre-trained to extract aesthetic attribute features and theme features, respectively. Then, the first level Attribute-Theme Graph (ATG) is built to investigate the coupling relationship between visual attributes and image theme. Further, a flexible aesthetics network is introduced to extract general aesthetic features, based on which we built the second level Attribute-Aesthetics Graph (AAG) to mine the relationship between theme-aware visual attributes and aesthetic features, producing the final aesthetic prediction. Extensive experiments on four public IAA databases demonstrate the superiority of the proposed TAVAR model over the state-of-the-arts. Furthermore, TAVAR features better explainability due to the use of visual attributes.
... Consequently, due to its multifaceted nature, researchers have employed various approaches, both objective and subjective, to measure creativity. On the one hand, studies have focused on objective formal-perceptual and design-oriented features, such as the Golden Ratio and luminance, comparing them to subjective preference ratings using classical and statistical learning methods [18][19][20][21] . Computational aesthetics has also evaluated aesthetic measures and quantification to generate designs (e.g. ...
... With the prediction of creativity judgements ratings as a target of art-attributes, we introduce a comprehensive method and a newly established initial model for art judgment analysis. The introduction of machine learning into art research is a consequent development of various methods to identify predictors of aesthetic appreciation and art evaluations, beginning with pure correlation or regression analyses [27][28][29] and more recent approaches using machine learning with objective values 8,[19][20][21] . ...
Creativity is a compelling yet elusive phenomenon, especially when manifested in visual art, where its evaluation is often a subjective and complex process. Understanding how individuals judge creativity in visual art is a particularly intriguing question. Conventional linear approaches often fail to capture the intricate nature of human behavior underlying such judgments. Therefore, in this study, we employed interpretable machine learning to probe complex associations between 17 subjective art-attributes and creativity judgments across a diverse range of artworks. A cohort of 78 non-art expert participants assessed 54 artworks varying in styles and motifs. The applied Random Forests regressor models accounted for 30% of the variability in creativity judgments given our set of art-attributes. Our analyses revealed symbolism, emotionality, and imaginativeness as the primary attributes influencing creativity judgments. Abstractness, valence, and complexity also had an impact, albeit to a lesser degree. Notably, we observed non-linearity in the relationship between art-attribute scores and creativity judgments, indicating that changes in art-attributes did not consistently correspond to changes in creativity judgments. Employing statistical learning, this investigation presents the first attribute-integrating quantitative model of factors that contribute to creativity judgments in visual art among novice raters. Our research represents a significant stride forward building the groundwork for first causal models for future investigations in art and creativity research and offering implications for diverse practical applications. Beyond enhancing comprehension of the intricate interplay and specificity of attributes used in evaluating creativity, this work introduces machine learning as an innovative approach in the field of subjective judgment.
... Such variability is even found for very basic SIPs. For example, Mallon et al. (2014) compute Aspect Ratio as height by width while Li et al. (2006) and Iigaya et al. (2021) calculate it as width by height. ...
... In the field of aesthetic research, they are also getting more and more attention recently. Standard CNNs trained for object recognition can be used to predict aesthetic judgments or to discriminate art from non-art, even though they have never been trained for these tasks , Iigaya et al. 2021, Conwell et al. 2022). All approaches have in common that they extract learned features from internal representations of neuronal networks. ...
... By contrast, SIPs that reflect the distribution of color or luminance gradients have a relatively clear relationship to the function of human visual system. A recent study by Iigaya et al. (2021) found that low-level (objective) SIPs such as hue or saturation are more strongly represented in the lower layers of CNNs, while high-level (subjective) image features such as concreteness or dynamics tend to be found in the upper layers. To what extent the information captured by hand-crafted SIPs and the features learned by neural networks overlap, remains to be elucidated. ...
In the fields of Experimental and Computational Aesthetics, numerous image datasets have been created over the last two decades. In the present work, we provide a comparative overview of twelve image datasets that include aesthetic ratings (beauty, liking or aesthetic quality) and investigate the reproducibility of results across different datasets. Specifically, we examine how consistently the ratings can be predicted by using either (A) a set of 20 previously studied statistical image properties, or (B) the layers of a convolutional neural network developed for object recognition. Our findings reveal substantial variation in the predictability of aesthetic ratings across the different datasets. However, consistent similarities were found for datasets containing either photographs or paintings, suggesting different relevant features in the aesthetic evaluation of these two image genres. To our surprise, statistical image properties and the convolutional neural network predict aesthetic ratings with similar accuracy, highlighting a significant overlap in the image information captured by the two methods. Nevertheless, the discrepancies between the datasets call into question the generalizability of previous research findings on single datasets. Our study underscores the importance of considering multiple datasets to improve the validity and generalizability of research results in the fields of experimental and computational aesthetics.
... It can be seen that the difference between realistic and abstract images is noticeable. Iigaya's study (2021) [35] on natural human behavior introduced four image characteristics, including whether the image is abstract or realistic and exhibiting a preference for abstract art. Previous studies have explored that pleasure during viewing is first enhanced by arousal of factors such as image novelty, com-plexity, unfamiliarity, etc., but then decreases as arousal (especially complexity) becomes too strong [36,37]. ...
The design of the light environment plays a critical role in the interaction between people and visual objects in space. Adjusting the space’s light environment to regulate emotional experience is more practical for the observers under lighting conditions. Although lighting plays a vital role in spatial design, the effects of colored lights on individuals’ emotional experiences are still unclear. This study combined physiological signal (galvanic skin response (GSR) and electrocardiography (ECG)) measurements and subjective assessments to detect the changes in the mood states of observers under four sets of lighting conditions (green, blue, red, and yellow). At the same time, two sets of abstract and realistic images were designed to discuss the relationship between light and visual objects and their influence on individuals’ impressions. The results showed that different light colors significantly affected mood, with red light having the most substantial emotional arousal, then blue and green. In addition, GSR and ECG measurements were significantly correlated with impressions evaluation results of interest, comprehension, imagination, and feelings in subjective evaluation. Therefore, this study explores the feasibility of combining the measurement of GSR and ECG signals with subjective evaluations as an experimental method of light, mood, and impressions, which provided empirical evidence for regulating individuals’ emotional experiences.
... Some of these are objective features, having to do with an artwork itself. For example, Iigaya et al. (2021) identify 13 visual features of abstract and impressionist art. For poetry, such objective characteristics obviously will differ, including, for example, assonance, pitch contours, and meter (Sharinger et al., 2022). ...
... Mathematical modeling can demonstrate how integration of such disparate vectors of aesthetic pleasure or displeasure is possible. Iigaya et al. (2021) use a linear summation that integrates 13 features to generate a predictive model of the preferences individual viewers, on average, show for a set of impressionist and abstract artworks. They identified 9 low-level visual features (e.g., brightness contrast, width/height ratio, blurring, and the presence of a human figure), as well as 4 higher-order features (abstract/concrete, dynamic/still, hot/cold, and positive/negative emotion). ...
... However, the average ratings given by all observers are better at predicting preference than their model; this suggests the importance of other factors and/or processes in taste. In other words, even though they cannot identify all of the elements driving preference, Iigaya et al. (2021) convincingly show that the human brain generates aesthetic preferences at least in part by summing over several component features, and it is thus possible to begin to understand how aesthetic judgments might integrate over a multidimensional preference space. Such a preference space has the potential for numerous, divergent outcomes, depending on an individual's weighting of each factor. ...
Aesthetic experiences have the potential to promote learning and creativity by enhancing the ability to understand complexity and to integrate novel or disparate information. Offering a theoretical framework for understanding the cognitive benefits of aesthetic experiences, this paper argues they are the necessary outcome of human learning, in which natural objects or artworks are evaluated in a multi-dimensional preference space shaped by Bayesian prediction. In addition, it contends that the brain-states underlying aesthetic experiences harness configurations of the apex three transmodal neural systems-the default mode network, the central executive network, and the salience network-that may offer information-processing advantages by recruiting the brain's high-power communication hubs, thus enhancing potential for learning gain.
... Although a rich variety of computational aesthetic metrics have been developed, these methods require prior human knowledge and hand-crafted feature extraction rules, which cannot fully cover the rich implicit features of the artworks. Recently, some researchers attempt to combine these metrics with deep neural networks [34]. But these studies are limited to predicting subjective ratings of human paintings. ...
... Computational aesthetics. In order to gain more interpretable insights into the similarity and difference between human and AI paintings, we design a number of computational aesthetic features based on previous studies [19,34,38,43] and our own knowledge. The features include four aspects of aesthetics, as follows. ...
The recent advances of AI technology, particularly in AI-Generated Content (AIGC), have enabled everyone to easily generate beautiful paintings with simple text description. With the stunning quality of AI paintings, it is widely questioned whether there still exists difference between human and AI paintings and whether human artists will be replaced by AI. To answer these questions, we develop a computational framework combining neural latent space and aesthetics features with visual analytics to investigate the difference between human and AI paintings. First, with categorical comparison of human and AI painting collections, we find that AI artworks show distributional difference from human artworks in both latent space and some aesthetic features like strokes and sharpness, while in other aesthetic features like color and composition there is less difference. Second, with individual artist analysis of Picasso, we show human artists' strength in evolving new styles compared to AI. Our findings provide concrete evidence for the existing discrepancies between human and AI paintings and further suggest improvements of AI art with more consideration of aesthetics and human artists' involvement.
