14:00 – 15:00 – Audrey Repetti (Heriot-Watt University)

Title: An optimisation view of learning robust and flexible denoisers for inverse imaging problems

Abstract: Proximal methods have been extensively used for solving inverse imaging problems. In this context, they are used to find an estimate of an unknown image from degraded measurements, by solving a regularised variational problem. Recently, proximal methods have been mixed with learning approaches to further improve the reconstruction quality. In particular, several works have proposed to replace the operator related to the regularisation by a more sophisticated denoiser, leading to plug-and-play (PnP) methods. In this presentation, we will discuss how optimisation theory can be leveraged to build robust and flexible denoisers. We will show that the resulting PnP methods are competitive with state-of-the-art methods for solving inverse imaging problems.

Refreshments available between 15:00 – 15:30, Huxley Common Room (HXLY 549)