Annonce en lien avec l’Action/le Réseau : Doctorants
Laboratoire/Entreprise : ICube UMR 7357, Université de Strasbourg, CNRS, France
Durée : 3 years
Contact : remi.allegre@unistra.fr
Date limite de publication : 2020-06-01
Contexte :
By-example texture synthesis aims at generating and editing visual textures similar to input texture samples, the former being larger than the latter, or of the same size depending on the applications. The two prominent approaches are procedural modeling and data-driven texture synthesis. Procedural approaches seek to represent textures by a mathematical model, whereas data-driven techniques rely on the matching of pixel neighborhoods or global image statistics on pixel intensities or filter responses. Inverse procedural modeling from arbitrary examples is a scientific bottleneck. It stems from a lack of image analysis tools tailored for the decomposition of the various visual texture features, especially structure and noise at multiple scales.
Sujet :
The goal of this thesis is to devise novel unsupervised or semi-supervised methods to learn visual texture features in the context of inverse procedural modeling and by-example texture synthesis. Particular attention will be paid to challenging heterogeneous textures exhibiting complex structures. The developed approach will be inspired by representation learning, with a focus on image factorization and attribute learning. Two fields of applications will be considered: texturing of virtual 3D environments and data augmentation for the training of models in histopathological slide image analysis.
Profil du candidat :
The position is offered to both foreign and French students who hold a Master degree in computer science.
Formation et compétences requises :
– Data science and/or computer graphics or image processing
– Basic skills in machine and deep learning
– Knowledge in texture synthesis is a plus
Adresse d’emploi :
ICube UMR 7357
300 bd Sébastien Brant
F-67412 Illkirch
France
Document attaché : Sujet_de_these_IA_Dischler_2020_EN.pdf