Selection du parametre de regularisation dans le probleme L2-Potts

J. Frecon, N. Pustelnik, N. Dobigeon, H. Wendt and P. Abry
Colloque francophone de traitement du signal et des images (GRETSI), 2017


The problem of choosing the regularization parameter involved in variational approaches for solving inverse problems remains an open question. In this study, we are interested in choosing this parameter, denoted lambda, for the L2-Potts model which is of interest for piecewise constant signal denoising. Its link with a hierarchical Bayesian model is highlighted, it allows a penalization over lambda to be derived and an algorithmic solution leads to an efficient estimation of lambda.


PDF   BibTex   Toolbox