Steerable Texture Synthesis

Abstract

Texture synthesis is typically concerned with the creation of an arbitrarily sized texture from a small sample, where the pattern of the generated texture should be perceived as resembling the example. Most of the current work follows a Markov model approach. The texture is generated by finding best matching pixels or patches in the sample and then copying them to the target. Here we extend this concept to incorporate arbitrary filters acting on the sample before matching and transferring. The filters may vary over the generated texture. Steering the filters with properties connected to the output image allows generating a variety of effects.