Parallel composite texture synthesis

Many textures require complex models to describe their intricate structures. Their modeling can be simplified if they are considered composites of simpler subtextures. After an initial, unsupervised segmentation of the composite texture into the subtextures, it can be described at two levels. One is a label map texture, which captures the layout of the different subtextures. The other consists of the different subtextures. This scheme has to be refined to also include mutual influences between textures, mainly found near their boundaries. The proposed composite texture model also includes these. The paper describes an improved implementation of this idea. Whereas in a previous implementation subtextures and their interactions were synthesized sequentially, this paper proposes a parallel implementation, which yields better results with simpler models. HE intricate nature of many textures makes it difficult to extract models that are compact and that support high quality synthesis. Often, the problem can be reduced by considering the texture as a composite of simpler subtextures. We propose such hierarchical approach to texture synthesis. We show that this approach can be used to synthesize intricate textures and even complete scenes, and that it also improves the results for "simple" textures. This suggests that hierarchical approaches to texture synthesis hold good promise as a general principle. Before explaining the composite texture algorithm, we concisely describe our basic texture model for single textures, in order to make this paper more self-contained. The point of departure of the basic model is the co-occurrence principle. Simple statistics about the colors at pixel pairs are extracted, where the pixels take on carefully selected, relative positions. It differs in this selectivity from more broad-brush co- occurrence methods (3), (4). Every different type of pair - i.e. every different relative position - is referred at as a clique type. The statistics gathered for these cliques are the histograms of the intensity differences between the head and tail pixels of the pairs, and this for all three color bands R, G, and B. Hence, the basic model consists of a selection of