Inpainting, which refers to the process of inferring unknown parts of visual content, such as images, videos or surfaces, has witnessed spectacular progresses over the last 20 years, and has been a very fruitful stimulation for the mathematical modeling of such visual content. In this symposium, we aim at bringing together new models and new application fields, as well as stimulating exchanges between different scientific communities. Specifically, the symposium will address sparse inpainting for compression, convolutional neural networks for semantic inpainting, motion-consistent video inpainting, and surfaces interpolation.

Sparse Inpainting with Anisotropic Integrodifferential Operators

Joachim Weickert (Saarland University)

Image Completion by CNN with Global and Local Consistency

Hiroshi Ishikawa (Waseda University)

Higher-order total directional variation with imaging applications

Simone Parisotto (University of Cambridge)

Motion Consistent Video Inpainting

Andrés Almansa (MAP5 - CNRS - Université Paris Descartes)

Organizers:

Yann Gousseau (Telecom ParisTech)

Simon Masnou (Université Lyon 1)

Keywords:

deep learning, image compression, image enhancement, image reconstruction, image representation, inpainting, partial differential equation models