Duality-based splitting for fast ℓ¹-TV image restoration
Christian Clason, Bangti Jin, and Karl Kunisch
We consider the problem of restoring images degraded by blurring and impulsive noise (e.g. salt-and-pepper noise). For such noise models, an ℓ¹ data-fitting term is more robust than standard ℓ² terms, while edges in the image are better preserved by a total variation penalty. The aim is to develop fast methods for solving the resulting non-smooth minimization problem, which can be implemented on graphics processing units (GPUs).
