Editing consists in modifying image information or statistics. Most of editing approaches rely on patch-based models that are empirically optimized with greedy algorithms. It makes difficult the control of global statistics of the generated images. Optimal Transport (OT) is an alternative to optimize globally complex editing models. OT offers powerful and flexible tools to interpolate between statistics in large scale problems. It is nevertheless required to adapt OT models to define efficient color editing methods. This minisymposium thus focuses on recent works mixing patches and OT to deal with color transfer, style transfer, texture synthesis and image interpolation.

Inverse problems using regularized optimal transport for computer graphics

Nicolas Bonneel (CNRS/LIRIS)

Image and Video Unsupervised Style Transfer by Adaptive Patch Sampling

Neus Sabater (Technicolor Research & Innovation)

Semi-Discrete Optimal Transport in Patch Space for Enriching Gaussian Textures

Arthur Leclaire (CMLA, ENS Cachan)

Generalized Optimal Transport for Manifold-Valued Images.

Friederike Laus (Technische Universität Kaiserslautern)

Organizers:

Rabin Julien (CNRS, Normandie Univ.)

Nicolas Papadakis (CNRS, Institut de Mathématiques de Bordeaux)

Keywords:

computer graphics, image enhancement, optimal transport