Coevolving Artistic Images Using OMNIREP

TL;DR

This paper addresses automatic generation of artistic images by coevolving a representation that encodes image positions and an interpreter that renders them into shapes. Using the OMNIREP cooperative coevolution framework, two interacting populations are evolved to optimize image similarity to inspiration images while allowing creative variation. Three rendering setups—chunks, polygons, and circles—demonstrate the framework's flexibility, with fitness based on mean absolute error against the inspiration images. The work establishes OMNIREP as a general meta-algorithm for discovering effective representation–interpreter pairings in creative domains, and points to future enhancements such as expanded color palettes, additional shapes, and richer interpreters to broaden artistic diversity and applicability.

Abstract

We have recently developed OMNIREP, a coevolutionary algorithm to discover both a representation and an interpreter that solve a particular problem of interest. Herein, we demonstrate that the OMNIREP framework can be successfully applied within the field of evolutionary art. Specifically, we coevolve representations that encode image position, alongside interpreters that transform these positions into one of three pre-defined shapes (chunks, polygons, or circles) of varying size, shape, and color. We showcase a sampling of the unique image variations produced by this approach.

Figures (7)

• Figure 1: Coevolution: (a) cooperative: Purple-throated carib feeding from and pollinating a flower (credit: Charles J Sharp, https://commons.wikimedia.org/wiki/File:Purple-throated_carib_hummingbird_feeding.jpg); (b) competitive: predator and prey---a leopard killing a bushbuck (credit: NJR ZA, https://commons.wikimedia.org/wiki/File:Leopard_kill_-_KNP_-_001.jpg); (c) commensalistic: Phoretic mites attach themselves to a fly for transport (credit: Alvesgaspar, https://en.wikipedia.org/wiki/File:Fly_June_2008-2.jpg).
• Figure 2: OMNIREP algorithm adapted to the task of evolutionary art. (A). OMNIREP includes two coevolving populations, one with candidate representations, and the other with candidate interpreters. Each population is evolved using the same fundamental evolutionary algorithm mechanisms (as summarized in the purple boxes). (B) The fitness of a given representation depends on representative interpreters (i.e., the 4 interpreters with the best fitness from the previous generation). In this example $R_{3}$'s fitness is the average fitness of the four representation-interpreter pairings. The fitness of a given interpreter (e.g., $I_{3}$) similarly depends on representative representations (not shown). The fitness of a representation-interpreter pair is computed by combining a representation individual (R) with an interpreter individual (I) to produce the pixel values of an image. Pair fitness is the mean absolute error between the pixels of the new image vs. the inspiration image.
• Figure 3: Overview and examples of the three image-mapping strategies employed by OMNIREP in experimental evolutionary runs.
• Figure 4: Inspirational images.
• Figure 5: Evolutionary trajectories of evolving images (chunks). Each row represents a single run.