Large block inpainting by color continuation analysis

Timothy K. Shih, Rong Chi Chang, Liang Chen Lu, Louis H. Lin

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

5 Scopus citations

Abstract

Automatic inpainting is a mechanism which repairs damaged pictures using an approximation mechanism. The most difficult problem is to inpaint a large damaged area, without knowing its content. One possible solution is to use color interpolation or extrapolation on surrounding pixels. However, spatial characteristics such as edges and pixel continuations are hard to be restored. In this research, we propose a series of automatic algorithms, which is based on an analysis of color continuations. Large damaged blocks are repaired, before the rest smaller potions are repaired by a multi-resolution inpainting algorithm. The mechanism is tested on more than 2000 images, including cartoon drawing, photos, Chinese painting, and western painting. Our results prove that, the proposed automatic mechanism fixes damaged image up to a certain degree of satisfaction from the users. The demonstration of our work is available at: http://www.mine.tku.edu.tw/demos/inpaint.

Original languageEnglish
Title of host publicationProceedings - 10th International Multimedia Modelling Conference, MMM 2004
EditorsY.-P.P. Chen
Pages196-202
Number of pages7
DOIs
StatePublished - 2004
EventProceedings - 10th International Multimedia Modelling Conference, MMM 2004 - Brisbana, Australia
Duration: 5 Jan 20047 Jan 2004

Publication series

NameProceedings - 10th International Multimedia Modelling Conference, MMM 2004

Conference

ConferenceProceedings - 10th International Multimedia Modelling Conference, MMM 2004
Country/TerritoryAustralia
CityBrisbana
Period5/01/047/01/04

Keywords

  • Digital inpainting
  • Image restoration
  • Large block inpainting
  • Multi-resolution inpainting

Fingerprint

Dive into the research topics of 'Large block inpainting by color continuation analysis'. Together they form a unique fingerprint.

Cite this