Graph-based text segmentation using a selected channel image

Chao Zeng, Wenjing Jia, Xiangjian He, Jie Yang

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

Abstract

This paper proposes a graph-based method for segmentation of a text image using a selected colour-channel image. The text colour information usually presents a two-polarity trend. According to the observation that the histogram distributions of the respective colour channel images are usually different from each other, we select the colour channel image with the histogram having the biggest distance between the two main peaks, which represents the main foreground colour strength and background colour strength respectively. The peak distance is estimated by the mean-shift procedure performed on each individual channel image. Then, a graph model is constructed on a selected channel image to segment the text image into foreground and background. The proposed method is tested on a public database, and its effectiveness is demonstrated by the experimental results.

Original languageEnglish
Title of host publicationProceedings - 2010 Digital Image Computing
Subtitle of host publicationTechniques and Applications, DICTA 2010
Pages535-539
Number of pages5
DOIs
Publication statusPublished - 2010
Externally publishedYes
EventInternational Conference on Digital Image Computing: Techniques and Applications, DICTA 2010 - Sydney, NSW, Australia
Duration: 1 Dec 20103 Dec 2010

Publication series

NameProceedings - 2010 Digital Image Computing: Techniques and Applications, DICTA 2010

Conference

ConferenceInternational Conference on Digital Image Computing: Techniques and Applications, DICTA 2010
Country/TerritoryAustralia
CitySydney, NSW
Period1/12/103/12/10

Keywords

  • Colour channel image
  • Graph cut
  • Histogram
  • Mean-shift
  • Text segmentation

ASJC Scopus subject areas

  • Computational Theory and Mathematics
  • Computer Science Applications

Fingerprint

Dive into the research topics of 'Graph-based text segmentation using a selected channel image'. Together they form a unique fingerprint.

Cite this