Learning optimal orders of the underlying Euclidean norm in total variation image denoising

Abstract

A novel class of semi-norms, generalising the notion of the isotropic total variation $T{V}_2$ and the an-isotropic total variation $T{V}_1$ is introduced. A supervised learning method via bilevel optimisation is proposed for the computation of optimal parameters for this class of regularizers. Existence of solutions to the bilevel optimisation approach is proven. Moreover, a finite-dimensional approximation scheme for the bilevel optimisation approach is introduced that can numerically compute a global optimizer to any given accuracy.

Cited by in corpus (1)

• Parameter learning and fractional differential operators: application in image regularization and decomposition