Synthesis-Oriented Texton

Presentation   -   Preprint   -   Source codes   -   Tutorial   -   Credits

Presentation

  The synthesis-oriented texton (SOT) is a small image that summarizes a Gaussian texture in a prescribed compact support, and that can be used for efficient discrete spot noise synthesis. The discrete spot noise associated to the SOT is indeed a good approximation of the limiting Gaussian random field, even for low intensity, as can be seen on the example below. The construction of this texton was generalized to the continuous domain in the Texton Noise project in order to address procedural texture synthesis.

	*		=
SOT		Poisson Process		Discrete Spot Noise

The SOT is described in the paper

"A Texton for Fast and Flexible Gaussian Texture Synthesis" (Bruno Galerne, Arthur Leclaire, Lionel Moisan), proceedings of the European Signal Processing Conference (Eusipco), 2014
Revised Preprint. Published Version. Hal.

  The experiments shown in this paper can be reproduced using the codes available below. The spot noise synthesis using the SOT is also illustrated on this tutorial page.

Source codes

  In the archive sot_matlab.zip, one can find functions (see the list below) related to the SOT computation and spot noise synthesis, a script file article_experiments.m allowing to reproduce the experiments shown in the article, a script file tuto.m containing the code presented on the tutorial page, a script file additional_experiments.m allowing to run additional experiments, two data files vary_tex6_* (containing the results of expensive Monte-Carlo simulation), and texture samples used in those experiments.

  The functions can also be downloaded separately:

• SOT computation: so_texton.m, cov_correction.m.
• ADSN synthesis: adsn.m, adsn_periodic.m.
• DSN synthesis: dsn.m, dsn_periodic.m.
• RPN synthesis: rpn.m.
• Canonical/Luminance texton: canonical_texton.m, luminance_texton.m.
• L2 Optimal-transport distance between circular ADSN models: otdist.m.
• Miscellaneous: disc.m, maximize.m

  If you find some bugs or mistakes in those codes, you can report them by email to Arthur Leclaire (firstname.lastname@parisdescartes.fr).

Credits

  The texture samples were made available by researchers whom I would like to thank warmly.

 The examples with name c*.png come from Lionel Moisan's Gaussian textures database, which is attached to the article

A Texton for Random Phase and Gaussian Textures. (Agnès Desolneux, Lionel Moisan, and Samuel Ronsin), preprint MAP5 nº 2015-xx, 2015.

This article contains a detailed study of the canonical (or luminance) texton associated to a texture, and also discusses the color correction step implemented in the function cov_correction.m.

  The textures IMG_7396.png and IMG_7406.png come from Bruno Galerne's personal collection.

  The other examples were made available by the authors of the article

Gabor Noise by Example. (Bruno Galerne, Ares Lagae, Sylvain Lefebvre and George Drettakis), ACM Transactions on Graphics (Proceedings of ACM SIGGRAPH 2012) 31(4), 2012. Webpage.