Adaptive space-time patch-based method for image sequence restoration

Jérôme Boulanger (1, 2) , Charles Kervrann (3, 2) , Patrick Bouthemy (2)

1 UR - Université de Rennes
2 VISTA - Vision spatio-temporelle et active
3 MIA-JOUY - Unité de biométrie et intelligence artificielle de Jouy

We present a novel space-time exemplar-based method for image sequence restoration. We have designed a locally adaptive estimation framework which relies on the analysis of the bias-variance trade-off. It allows us to select on-line the best adapted filtering space-time neigh-borhood at each pixel. Furthermore, it involves an original space-time exemplar-based restoration which improves the overall performance. This method is unsupervised and requires no motion estimation. It nevertheless can be combined with motion estimation to cope with very large displacements due to camera motion. Experiments show that this method is able to drastically improve the quality of highly corrupted image sequences. Quantitative results on standard image sequences demonstrate that our approach outperforms other recent methods.

Mots clés: removal; motion; image sequence restoration; nonparametric estimation; nonlinear filtering; bias-variance trade-off; deplacement

Mots clés: image restauration d'ordre; evaluation nonparamétrique; filtrage non-lineaire; mouvement; INTELLIGENCE ARTIFICIELLE; INFORMATIQUE

Domaines: Mathématiques [math]; Informatique [cs]

Jérôme Boulanger, Charles Kervrann, Patrick Bouthemy. Adaptive space-time patch-based method for image sequence restoration. SMVP'06 Statistical Methods in Multi-Image and Video Processing, May 2006, Graz, Austria. ⟨hal-02755988⟩