HAL will be down for maintenance from Friday, June 10 at 4pm through Monday, June 13 at 9am. More information
Skip to Main content Skip to Navigation
Habilitation à diriger des recherches

Patch-based models for image post-production

Abstract : The objective of images and videos post-production is to enhance visual quality, add details or restore old data. This document presents different patch-based post-production methods based on the self-similarity principle. After a brief summary of my research activities since 2008, a review of metrics used for computing the similarity between patches is presented. Their applications to image completion (inpainting) and grayscale image colorization are then discussed. Patch-based methods are one-path greedy algorithms: pixels are processed one after the others without being further modified. This property may lead to the well-known growing garbage problem. Another category of methods used to solve post-production problems, is made of geometric methods. They rely on the diffusion of colors obtained by minimizing functionals or solving PDE. Their iterative optimization lead to smooth results but fail at reconstructing small details and textures. This document describes several patch-based functionals designed for inpainting and colorization, that intent to combine both categories of methods. The corresponding results are generally realistic, with continuous contours and fine textures reconstruction.
Document type :
Habilitation à diriger des recherches
Complete list of metadata

Cited literature [206 references]  Display  Hide  Download

Contributor : Aurélie Bugeau Connect in order to contact the contributor
Submitted on : Monday, July 2, 2018 - 4:55:16 PM
Last modification on : Saturday, March 5, 2022 - 3:18:03 PM
Long-term archiving on: : Monday, October 1, 2018 - 9:32:58 AM


Files produced by the author(s)


  • HAL Id : tel-01811035, version 2



Aurélie Bugeau. Patch-based models for image post-production. Signal and Image Processing. Université de Bordeaux, 2018. ⟨tel-01811035v2⟩



Record views


Files downloads