Vector solid texture synthesis using unified RBF-based representation and optimization

Solid textures are essential for modeling virtual internal materials. Existing approaches either generate raster solid textures or only focus on vector representation. To facilitate efficient synthesis and intuitive editing of vector solid texture, we propose the novel solid texture representation, named radial basis function (RBF) solid texture. An RBF solid texture consists of a set of spatially distributed RBF instances. Each RBF instance encapsulates a 3D position, an RGB color and a signed distance field (SDF) value. Such a representation is resolution independent, compact in storage and capable of supporting efficient random access with an indexing uniform grid. We directly synthesize RBF solid texture from raster exemplar by minimizing an energy function, which encodes the position, color and SDF difference between output volumetric RBF instances and input example planar RBF instances. The minimization process iteratively updates output RBF instances with an EM algorithm. Our experiments show that our algorithm can produce RBF solid textures in high efficiency and compact storage for a variety of exemplars, including stochastic patterns or more structured patterns. Furthermore, RBF solid textures we proposed benefit intuitive editing for either region-based and RBF-based effects.