Universal image statistics of visual art as the basis of esthetic perception

• Antragsteller: Professor Dr. Christoph Redies
• Fachliche Zuordnung: Molekulare Biologie und Physiologie von Nerven- und Gliazellen
• Förderung: Förderung von 2007 bis 2012
• Projektkennung: Deutsche Forschungsgemeinschaft (DFG) - Projektnummer 51810560

One of the hypotheses of experimental (neuro-)aesthetics is that artists create their works of art with specific properties that elicit an aesthetic response in humans. Specifically, we proposed that these properties reflect basic functional properties of the human visual system, such as the efficient sensory coding of natural images. This idea is based on the finding that aesthetic artworks and natural scenes share fractal-like (scale-invariant) statistical properties in the Fourier domain. Other aesthetic images, such as graphic novels, also display this property. In contrast, non-aesthetic man-made images, such regular text, deviate from scale-invariant statistics. To continue our project, we propose to investigate other statistical properties that can distinguish aesthetic and non-aesthetic images. A particular focus is on algorithms that combine local with global features, such as self-similarity. Other mechanisms to be studied are direct feature-based analysis, semantic segmentation and color features. In accordance with the first funding period, we continue work at three different levels: (1) investigation and implementation of feature extraction mechanisms, (2) modeling of the statistical properties of images based on the extracted feature, and (3) a feed-back mechanism to compare our developed models with the performance of the human visual system. This feedback will be achieved by creating synthetic images that will be evaluated by human observers.

• DFG-Verfahren: Sachbeihilfen
• Beteiligte Person: Professor Dr.-Ing. Joachim Denzler