Bayesian rendering with non-parametric multiscale prior model

This paper investigates the use of the Bayesian inference for devising an example-based rendering procedure. As a prior model of this Bayesian inference, we exploit the multiscale non-parametric model recently proposed by Wei et al. (2000) for texture synthesis. This model appears to be interesting also to capture some characteristics of a rendering style from an artistic illustration example. The results obtained, with a prior model capturing the rendering style of drawing samples or trained with synthetic and real input textures, are presented. They indicate that the proposed method allows us to simulate automatic synthesis of various illustration styles. More generally, the proposed scheme is able to re-render an input image in the style of another image allowing the creation of a very broad range of artistic and visual effects.