Assessing aesthetics of generated abstract images using correlation structure

TL;DR

This paper investigates whether the aesthetic quality of abstract images can be characterized by their correlation functions, showing that architecture influences these functions and that aesthetic images exhibit distinct correlation patterns.

Contribution

The study demonstrates that correlation functions are influenced by network architecture and are indicative of aesthetic quality in generated abstract images.

Findings

Correlation functions are largely determined by network architecture.

Aesthetic images have distinct correlation function patterns.

Human judgment aligns with correlation-based distinctions.

Abstract

Can we generate abstract aesthetic images without bias from natural or human selected image corpi? Are aesthetic images singled out in their correlation functions? In this paper we give answers to these and more questions. We generate images using compositional pattern-producing networks with random weights and varying architecture. We demonstrate that even with the randomly selected weights the correlation functions remain largely determined by the network architecture. In a controlled experiment, human subjects picked aesthetic images out of a large dataset of all generated images. Statistical analysis reveals that the correlation function is indeed different for aesthetic images.