Accueil > Publications > Thèses du CMLA
Recherche - Valorisation
Exemplar based texture synthesis: models and applications
le 3 octobre 2016
9h00
Thèse de doctorat de Lara Raad (CMLA)
lara.jpg
Exemplar based texture synthesis is the process of generating, from an input texture sample, new texture images that are perceptually equivalent to the input. There are roughly two main categories of algorithms: the statistics-based methods and the non-parametric patch-based methods.
The statistics-based methods aim at characterizing a given texture sample by estimating a set of statistics which will define an underlying stochastic process. The new images will then be samples of this stochastic process, i.e. they will have the same statistics as the input sample. The question here is what would be the appropriate set of statistics to yield a correct synthesis for the wide variety of texture images? The results of statistical methods are satisfying but only on a small group of textures, and often fail when important structures are visible in the input.
The non-parametric patch-based methods reorganize local neighborhoods from the
input sample in a consistent way to create new texture images. These methods return impressive visual results. Nevertheless, they often yield verbatim copies of large parts of the input sample. Furthermore, they can diverge, starting to reproduce iteratively one part of the input sample and neglecting the rest of it, thus growing what experts call garbage.
In this thesis we propose a technique combining ideas from the statistical based methods and from the non-parametric patch-based methods. We call it the locally Gaussian method. The method keeps the positive aspects of both categories: the innovation capacity of the parametric methods and the ability to synthesize highly structured textures of the non-parametric methods. To this aim, the self-similarities of a given input texture are modeled with conditional multivariate Gaussian distributions in the patch space. A new image is generated patch-wise, where for each given patch a multivariate Gaussian model is inferred from its nearest neighbors in the patch space of the input sample, and hereafter sampled from this model. The synthesized textures are therefore everywhere different from the original.
In general, the results obtained are visually superior to those obtained with statistical based methods, which is explainable as we use a local parametric model instead of a global one. On the other hand, our results are comparable to the visual results obtained with the non-parametric patch-based methods. This dissertation addresses another weakness of patch-based methods. They are strongly dependent on the patch size, which has to be decided manually. It is therefore crucial to fix a correct patch size for each synthesis. Since texture images have, in general, details at different scales, we extend the method to a multiscale approach which reduces the strong dependency of the method on the patch size.
Patch based methods involve a stitching step. Indeed, the patches used for the synthesis process overlap each other. This overlap must be taken into account to avoid any transition artifact from patch to patch. Our first attempt to deal with it was to consider directly the overlap constraints in the local parametric model. The experiments show that for periodic and pseudo-periodic textures, considering these constraints in the parametrization is enough to avoid the stitching step. Nevertheless, for more complex textures it is not sufficient. This leads us to suggest a new stitching technique inspired by optimal transport and midway histogram equalization.
This thesis ends with an extensive analysis of the generation of several natural textures. This study shows that, in spite of remarkable progress for local textures, the methods proposed in the extensive literature of exemplar based texture synthesis are still limited when dealing with complex and non-stationary textures.
- Type :
- Thèses - HDR
- Lieu(x) :
- Campus de Cachan
Bâtiment Laplace, Salle Renaudeau, R-de-Ch.
Tutelle
Jury de thèse
Advisors
Agnès Desolneux
Jean-Michel Morel
Referees
Jean-François Aujol
Gabriel Peyré
Javier Portilla
Examiner
Yann Gousseau
Pablo Musé
Agnès Desolneux
Jean-Michel Morel
Referees
Jean-François Aujol
Gabriel Peyré
Javier Portilla
Examiner
Yann Gousseau
Pablo Musé