Session 4

Automatic Analysis of Emotions Conveyed by Abstract Painting

Victoria Yanulevskaya, Elia Bruni, Jasper R.R. Uijlings, Andreza Sartori, Elisa Zamboni, Francesca Bacci, David Melcher and Nicu Sebe

Last century, the painters of the abstract art movement employed basic visual features such as colours, shapes, and texture to convey emotions. We investigate which elements of the painting are associated with positive and negative emotions by analysing 500 abstract paintings from the collections of Modern Art Museum (MART), Rovereto. Specifically, we concentrate on paintings by Carlo Belli, Aldo Schmid, Luigi Senesi, and Luigi Veronesi, as their work is not the result of an improvisation but it is part of a deep theoretical reflection on the elements that make a painting. To obtain the ground truth, each painting was scored by 20 people as carrying a positive or negative response on a Liekert scale of 1-7. We use the state-of-the-art Bag-of-Words classification framework [Sivic and Zisserman, 2003, International Conference on Computer VIsion, 2, 1470-1477] to automatically predict emotional conveyed by the artwork and obtain an accuracy of around 80%. Additionally, we are also able to visualize how each element of a painting contributes to the overall emotional impression. This allows answering the important question why a specific painting is perceived as positive or negative. Many of the features identified by the computational analysis conform to principles outlined by the artists.

Fractals, scale-invariance and visual preference

Branka Spehar and Richard Taylor

At present, there exists a set of suggestive links between aesthetic preference, the fractal properties of images, and the statistics of natural scenes. We have previously shown that humans display a consistent preference for a certain range of fractal dimension across fractal images of various types (Spehar et al., 2003, Computer & Graphics, 27, 813-820). While some have suggested that fractal-like patterns are inherently pleasing because they resemble natural patterns and scenes, the relation between aesthetics, scale invariance of natural scenes and fractals (especially those defined as binary boundaries) remains unclear. Here we examine visual preference for 1/f noise images of varying slopes and compare these directly to preferences measured for a comparison set of thresholded (black and white) images. We found no significant differences in preferences between gray-scale images and binary comparison images obtained by simply thresholding the original gray-scale images. For both set of images, the visual preference peaked for images with the amplitude spectrum slopes from 1.25 to 1.5, thus confirming and extending the previously observed relationship between fractal characteristics of images and visual preference.

Kafka's Castle: Vision and Imagination in Visual Art and Literature

Emily Troscianko

This paper shows how literature can provide a context for drawing connections between visual perception, imagination, and visual art. Visual perception is an important element of literary art and the experiences it induces in readers: the ways in which characters are described as perceiving the fictional world have significant effects on readers' imaginative responses to this world (see e.g. Troscianko (ET), 2010, Language and Literature, 19, 151-171). I describe an experimental paradigm in which I evaluate experiences induced by the opening of Kafka's The Castle by 1) using a simple online measure of 'presence' (see e.g. Troscianko (T) and Hinde, 2011, i-Perception 2, 216 (in press)) and 2) asking participants to draw what they had imagined while reading. The results enrich the connections that can be drawn between specific approaches within vision science, notably the sensorimotor account of vision and visual consciousness (e.g. O'Regan and Noë, 2001, Behavioral and Brain Sciences, 24, 939-1031), and theories/concepts relating to visual detail and perspective in visual and verbal art. The empirical convergence of visual and verbal art helps us tease out distinctions between imagining, seeing, and conceptualising seeing, and suggests further avenues for exploring how vision acts as a mediator of aesthetic experience.

Aspects of experience of beauty

Slobodan Markovic

In a previous study we specified three dimensions of the subjective experience of visual stimuli: Hedonic Tone, Regularity and Arousal (Markovic and Jankovic, 2001, Perception, 30. ECVP Supplement, 30). These dimensions included descriptors which expressed different aspects of the experience of beauty: Hedonic Tone (descriptor: pleasant), Regularity (harmonious) and Arousal (interesting). The purpose of the present study was to specify the relationship between the judgment of beauty and judgments on the other three descriptors. Participants judged sixteen achromatic abstract visual patterns on four seven-step bipolar scales: beautiful-ugly, pleasant-unpleasant, harmonious-disharmonious and interesting-boring. The multiple regression indicated that the judgment of beauty was closer to judgments of pleasure than to judgments of harmony and interestingness. Inter-corelations suggested a model with one central cluster which included beauty and pleasure (correlations about 0.7), and two lateral descriptors, harmony and interestingness (correlations with descriptors of the central cluster were about 0.5); the correlation of harmony and interestingness was not significant. These results suggest that the concept of beauty can be reduced to pleasure, but that it includes two relatively independent aspects - harmony (beauty as figural goodness) and interestingness (beauty as impressiveness).

Goodness-of-fit of oriented elements within a rectangular frame

Stefano Guidi and Stephen E Palmer

Previous research (Palmer & Guidi, 2011, Perception, 40, 1428-1446) has shown that the structure of a rectangular frame strongly influences the perceived goodness-of-fit of small circular probe positioned within it. The centre is consistently rated as the best position, followed by positions along the vertical, horizontal and local diagonal symmetry axes. In the three present experiments we investigated how an element's goodness-of-fit is influenced by the relationships between its structure and that of the frame. Fit-ratings of isosceles triangles at different positions and orientations, revealed strong orientational effects, especially when the probe's axis of symmetry aligned with the frame's axes of symmetry, and directional effects when they point into the frame and/or toward the right. Fit-ratings within rectangular frames at different orientations (0°, 45°, 90°, and 135°) showed that these orientational effects were more strongly driven by alignment with the rectangle's sides than with gravitational or retinal reference frames. When line segments or small ovals were used with a denser space sampling, fit-ratings for the less-strongly-oriented ovals were lower, and ratings tended to decrease along the angle bisectors with distance from the corners. The results are relevant to the empirical study of aesthetic response to images within rectangular frames.

Mirror Reversal of Artworks

Michael Forster and Helmut Leder

When you would ask an artist if it would make a difference whether his/her artwork is mirrored or not, he/she would be at least annoyed. However, previous research has shown that we cannot reliably distinguish between an original artwork and its mirrored version (enantiomorph). Based on these findings, we conducted a series of experiments further exploring the nature of mirror reversal in artworks. In a first experiment, we compared originals and enantiomorphs of both very familiar artworks and rather unfamiliar artworks for liking, familiarity, and reversal detection. In line with a fluency account, only famous originals had the tendency to be liked more than their enantiomorphs, whereas unfamiliar originals showed no effect of orientation. This effect however, was independent of reversal detection, which was around chance level for all the artworks, regardless of familiarity. In further experiments we address the effects of task variation on judgments of reversal and liking and the effects of different procedures on recognition of originals and enantiomorphs. Together these findings shed light on the impact of mirror reversal on preference for artworks and on the nature of mirror reversal in image perception and recognition in general.