Advanced Reaction-Diffusion Models for Texture Synthesis

Abstract

Since the introduction of texture synthesis using a reaction-diffusion model in the early 1990s, their use has not been widespread. This is likely due to both the difficulty in selecting parameters that result in stable, useful patterns as well as the computational costs of producing these patterns. In this paper we present techniques that help overcome the difficult parameter selection process that controls the pattern development. In addition, we expand the basic diffusion model to allow for shaped patterns. Finally, we show that it is possible to create multiple and oscillating patterns by coupling two reaction-diffusion systems together. These techniques have been implemented using both explicit and semi-implicit solutions on a CPU and GPU.We provide sample source code of both implementations online.