Section: New Results

Texture Synthesis

Participants : Sylvain Lefebvre, Bruno Jobard.

We continued investigating on Gabor Noise and considered fitting the parameters of our Gabor noise texturing technique from example images. This required a new formulation of our noise, allowing us to solve the problem as a basis pursuit denoising optimization. This is the result of a collaboration with the team REVES / Inria Sophia-Antipolis, the Katholieke Universiteit of Leuven and Université Paris Descartes. This work has been presented at the SIGGRAPH conference this year [8] .

Figure 2. Gabor Noise by Example.

We also revisited techniques for texture synthesis explicitly copying and assembling large patches of an example image to form a new texture. We accelerate this process through a parallel implementation which both optimizes for the shape of the patches and a deformation along their boundary to better match edges. This work is part of the PhD thesis of Anass Lasram and has been presented this year at the Eurographics/ ACM SIGGRAPH Symposium on High Performance Graphics, [19] .

Figure 3. Parallel patch–based texture synthesis.

We also studied ways of helping the user to select the parameters of procedural texture generators, by proposing two contributions :

• We studied how to summarize the appearances generated by complex procedural textures in a small preview image. The challenge is to capture the large variety of appearances despite a limited pixel space. We formulate the problem as a layout of high-dimensional samples in a regular grid, and optimize for it through a modified Self Organizing Map algorithm. This work is part of the PhD thesis of Anass Lasram, and is a collaboration with our industrial partner Allegorithmic. This work has been published this year in the journal "Computer Graphics Forum", [10] .

• The parameters of complex procedural textures are typically chosen through a slider-based interface. We augment this interface with preview images which predict how the texture will change when manipulating the slider. This greatly simplifies the process of choosing parameters for these textures. This work is part of the PhD thesis of Anass Lasram, and is a collaboration with our industrial partner Allegorithmic. This work has been published this year as EUROGRAPHICS short paper, [18] .

Figure 4. Scented Sliders for Procedural Textures.