We develop new methods to easily texture-map 3D models. This concerns both the production of highly detailed content, the association of this texture content with complex 3D geometry and parameterization techniques to compute texture-efficient U,V coordinates.
Texture generation: We proposed in [ACM SIGGRAPH 2010] a method to transform an image into a texture generator that can produce textures of arbitrary sizes. Our new texture synthesis approach casts synthesis as a shortest path problem in a graph describing the space of images that can be synthesized. Each path in the graph describes how to form a new image by cutting strips of the source image and reassembling them in a different order. Only the paths describing the result need to be stored in memory: synthesized textures are reconstructed at rendering time. We also study image generation techniques that use a 3D model to steer texture generation [Tech report].These two projects was done in cooperation with the REVES team.
Easy texturing: We proposed in [ACM I3D 2010] a new tool for interactive texture assignment. Each time an image is chosen for a surface, our algorithm propagates this information throughout the entire environment. Our approach is based on a new surface similarity measure. We exploit this measure in an algorithm ranking all possible textures for a given surface. Hence, we do not simply assign a texture to the surface but also propose an ordered list of choices for the user. This project was done in cooperation with the REVES team. We also study an approach based on 3D texture, where the third axis is used to represent material variations [Computer Graphics Forum 2009] (in cooperation with Georgia Tech).
U,V mapping: Based on our work in geometry processing, we proposed geometry textures [Sibgrapi 2007], [Computer Graphics Forum 2008], a technique to encode fine-scale geometry in elevation maps, rendered by a fragment shader. We also studied the problems of seams, that appear when constructing a texture atlas, and proposed a method to completely eliminate them, based on a grid-preserving parameterization combined with an algorithm that computes the color constraints required to ensure continuity [EGSR 2010].
* State of the art in Procedural Noise Functions, Eurographics State of the art report, 2010
* Tutorial: (with source code): Parallel Controllable Texture Synthesis