Skip to Main content Skip to Navigation
Theses

Apprentissage de dictionnaires structurés pour la modélisation parcimonieuse des signaux multicanaux

Sylvain Lesage 1
1 METISS - Speech and sound data modeling and processing
IRISA - Institut de Recherche en Informatique et Systèmes Aléatoires, Inria Rennes – Bretagne Atlantique
Abstract : Sparse decompositions describe a signal as the combination of a few basis waveforms, called atoms. The dictionary of atoms, crucial for an efficient decomposition, may result from a prior choice (wavelets, Gabor atoms, ...) that fixes the dictionary structure or from a learning process on representative samples of the signal. Here, we propose a hybrid framework combining some structural constraints with a learning approach. Such structured dictionaries lead to a better adaptation to the properties of the signal and enable the handling of large amounts of data. We expose the concepts and tools that support this approach, notably the adaptation of the Matching Pursuit and K-SVD algorithms to dictionaries composed of linearly deformable motifs, via an adjunction property. Moreover, we present results of monochannel and multichannel separation using the proposed framework.
Complete list of metadatas

https://tel.archives-ouvertes.fr/tel-00564061
Contributor : Rémi Gribonval <>
Submitted on : Monday, February 7, 2011 - 10:14:43 PM
Last modification on : Friday, July 10, 2020 - 4:21:06 PM
Long-term archiving on: : Sunday, May 8, 2011 - 3:51:40 AM

Identifiers

  • HAL Id : tel-00564061, version 1

Citation

Sylvain Lesage. Apprentissage de dictionnaires structurés pour la modélisation parcimonieuse des signaux multicanaux. Traitement du signal et de l'image [eess.SP]. Université Rennes 1, 2007. Français. ⟨tel-00564061⟩

Share

Metrics

Record views

601

Files downloads

1193