Text location for scene image with inherent features

  • Qiaoyu Sun*
  • , Yue Lu
  • *Corresponding author for this work

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

Abstract

Locating the text from nature scene images is a challenge problem for computer, because text is various. In this paper, the inherent characteristics of the text are employed to locate the text from complicated background in the scene images. The edge map of the input image is obtained by the Canny operator which is masked by the Sobel edge map. Then the connected components analysis is used to detect the potential characters. Thirdly, the characters in the same line will be merged into a candidate text region. Finally, the symmetry of edge direction and the stroke width of each candidate text region are calculated to verify if it is a true text region. The experimental results on various scene images have demonstrated that the proposed method is capable of effectively locating text regions of the nature scene image.

Original languageEnglish
Title of host publicationPattern Recognition - Chinese Conference, CCPR 2012, Proceedings
Pages522-529
Number of pages8
DOIs
StatePublished - 2012
Event2012 5th Chinese Conference on Pattern Recognition, CCPR 2012 - Beijing, China
Duration: 24 Sep 201226 Sep 2012

Publication series

NameCommunications in Computer and Information Science
Volume321 CCIS
ISSN (Print)1865-0929

Conference

Conference2012 5th Chinese Conference on Pattern Recognition, CCPR 2012
Country/TerritoryChina
CityBeijing
Period24/09/1226/09/12

Keywords

  • Connected Component Analysis
  • Edge Detection
  • Stroke Width
  • Symmetry of Edge Direction
  • Text Location

Fingerprint

Dive into the research topics of 'Text location for scene image with inherent features'. Together they form a unique fingerprint.

Cite this