Skip to Main content Skip to Navigation

Probabilistic Approaches for the Digital Restoration of Television Archives

Abstract : Within the context of archives restoration, we investigate in this thesis the concealment of impulsive defects (blotches, video dropouts). Existing detection and correction methods reach their limits with the presence of motion estimation failures due to complex natural events. We aim at taking into account these events that we shall call pathological motion. For both detection and correction steps, we investigate probabilistic approaches and our algorithms are expressed by means of parametric or non-parametric Markov random fields. The proposed detection method relies on the framework of the Bayesian theory of estimation. We consider a larger temporal window than the usual three frames, in order to better distinguish defects from pathological motion and thus dramatically reduce the number of false alarms. We also propose a method to correct missing data areas which is inspired by works on texture synthesis. After generalizing these techniques to natural images, we integrate them in a spatio-temporal context which allows an implicit fallback on a spatial correction when motion is too complex. We first validate the proposed methods separately before combining them to create a complete prototype for the concealment of impulsive defects.
Complete list of metadatas

Cited literature [63 references]  Display  Hide  Download
Contributor : Raphaël Bornard <>
Submitted on : Saturday, January 7, 2012 - 8:44:47 PM
Last modification on : Friday, October 23, 2020 - 4:38:34 PM
Long-term archiving on: : Sunday, April 8, 2012 - 2:25:14 AM


  • HAL Id : tel-00657636, version 1



Raphaël Bornard. Probabilistic Approaches for the Digital Restoration of Television Archives. Signal and Image processing. Ecole Centrale Paris, 2002. English. ⟨tel-00657636⟩



Record views


Files downloads