Project Title - PHD 2022

Semi-Procedural Texture Model. Separation of procedural structure synthesis from per-pixel color synthesis. Starting with an example texture and its structure map (manual segmentation), we determine the closest procedural model (parameter estimation) and then its structure by thresholding. Next comes the colorization phase (post-processing).

Abstract

Automating the tasks of authoring textures and materials for 3D virtual worlds is in high demand in the film and video game industry.

In this context, we are interested in stochastic structured textures (and by extension to materials), among the most widespread. Characterized by spatially varying properties, they are particularly difficult to reproduce by by-example synthesis methods and have been the subject of little research work.

We propose a new method of textures and materials synthesis, called semi-procedural, consisting in separating the synthesis of the structure, from that of the color details. The structure, procedural, uses a new generic sparse convolutional noise function that is infinite, repeatless, coherent, editable and controllable. The details are generated by-example via an automatic parallel optimization approach resulting in visual similarity. This method takes advantage of the best world of procedural and by-example synthesis: results faithful to examples, low storage cost representation, ability to synthesize on the fly, high degree of control, and scalability.

Keywords: computer graphics, texture synthesis, procedural generation, noise functions, texture basis function, material, inverse procedural modeling

Semi-Procedural Appearance Model for Controlling and Scaling Texture and Material Synthesis

Abstract

Resources (Work In Progress: not finished yet!)

Related Project(s)