Automation of Fabric Pattern Construction using Genetic Algorithms

Abstract

This paper introduces the use of Genetic Algorithms to evolve fabric patterns from randomly generated seeds. The patterns are evolved from random, often dull coloring of the image, to bright multi-color patterns that are aesthetically pleasing in nature. The main problem that this paper intends to solve is to introduce complete automation in the design process of patterns, which have historically been dependent upon human arbitrators to judge the quality of intermediate outputs. In its stead, the proposed algorithm evaluates the quality of the image using inherent latent features present in the image itself. Our algorithm takes into account the distribution of color, global contrast, and the overall dullness score of the image to evaluate the quality of the generated patterns. To create diverse patterns that feel more natural, different approaches are experimented with. These include the use of L-systems and image processing techniques, in a bid to construct a pattern which seems more human-like, rather than just rudimentary digital art.