Texture synthesis using label assignment over a graph

Partition of unity parametrics (PUPs) are a recent framework designed for geometric modeling. We propose employing PUPs for procedural texture synthesis, taking advantage of the framework’s guarantees of high continuity and local support. Using PUPs to interpolate among data values distributed through the plane, the problem of texture synthesis can be approached from the perspective of point placement and attribute assignment; a graph over the points, such as the Delaunay triangulation, allows us to apply attributes or labels to the nodes in a structured way. We present several alternative mechanisms for point distribution and demonstrate how the system is able to produce a variety of distinct classes of texture, including analogs to cellular texture, Perlin noise, and progressively-variant textures. We further describe structured textures obtained by assigning label clusters using queries over the graph, such as breadth-first or depth-first traversal.