The best of both worlds: synthesis-based acceleration for physics-driven cosparse regularization

Srdan Kitic (1) , Nancy Bertin (1) , Rémi Gribonval (1)

1 PANAMA - Parcimonie et Nouveaux Algorithmes pour le Signal et la Modélisation Audio

Recently, a regularization framework for ill-posed inverse problems governed by linear partial differential equations has been proposed. Despite nominal equivalence between sparse synthesis and sparse analysis regularization in this context , it was argued that the latter is preferable from computational point of view (especially for huge scale optimization problems arising in physics-driven settings). However, the synthesis-based optimization benefits from simple, but effective all-zero initialization, which is not straightforwardly applicable in the analysis case. In this work we propose a multiscale strategy that aims at exploiting computational advantages of both regularization approaches.

Domaines: Traitement du signal et de l'image [eess.SP]

Srdan Kitic, Nancy Bertin, Rémi Gribonval. The best of both worlds: synthesis-based acceleration for physics-driven cosparse regularization. iTwist 2016 - International Traveling Workshop on Interactions Between Sparse Models and Technology, Aug 2016, Aalborg, Denmark. ⟨hal-01329051⟩