On Video Completion: Line Scratch Detection in Films and Video Inpainting of Complex Scenes

Abstract : The inevitable degradation of visual content such as images and films leads to the goal of image and video restoration. In this thesis, we look at two specific restoration problems : the detection of line scratches in old films and the automatic completion of videos, or video inpainting as it is also known. Line scratches are caused when the film physically rubs against a mechanical part. This origin results in the specific characteristics of the defect, such as verticality and temporal persistence. We propose a detection algorithm based on the statistical approach known as a contrario methods. We also propose a temporal filtering step to remove false alarms present in the first detection step. Comparisons with previous work show improved recall and precision, and robustness with respect to the presence of noise and clutter in the film. The second part of the thesis concerns video inpainting. We propose an algorithm based on the minimisation of a patch-based functional of the video content. In this framework, we address the following problems : extremely high execution times, the correct handling of textures in the video and inpainting with moving cameras. We also address some convergence issues in a very simplified inpainting context.
Complete list of metadatas

Cited literature [128 references]  Display  Hide  Download

https://tel.archives-ouvertes.fr/tel-01136253
Contributor : Alasdair Newson <>
Submitted on : Thursday, March 26, 2015 - 6:39:11 PM
Last modification on : Saturday, March 28, 2015 - 1:11:50 AM
Long-term archiving on : Thursday, July 2, 2015 - 8:16:55 AM

Identifiers

  • HAL Id : tel-01136253, version 1

Collections

Citation

Alasdair Newson. On Video Completion: Line Scratch Detection in Films and Video Inpainting of Complex Scenes. Image Processing [eess.IV]. Telecom ParisTech, 2014. English. ⟨tel-01136253⟩

Share

Metrics

Record views

260

Files downloads

787