Versatile Text Extraction System for Text-to-Speech Reading Assistant Camera

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Wearable camera device translating the text in the scene into speech is one of the most anticipated devices for the visually-impaired. The users would probably want to read any text using such a device. Although various scene text extraction methods have been developed so far, the target objects are most often limited to simple signboards, small memos, etc. We propose a versatile scene text extraction method that can handle a wide variety of targets including complex signboards with many text lines. Experimental results show that our system runs at a video rate and can extract densely arranged text lines even with some distortion and shading. A locally-adaptive binarization technique contributes to the better quality of extracted text images.

Original languageEnglish
Title of host publicationAssistive Technology
Subtitle of host publicationBuilding Bridges
EditorsCecilia Sik-Lanyi, Peter Cudd, Klaus Miesenberger, Evert-Jan Hoogerwerf
PublisherIOS Press
Pages392-397
Number of pages6
ISBN (Electronic)9781614995654
DOIs
Publication statusPublished - 2015 Jan 1
Event13th European Conference on the Advancement of Assistive Technology, AAATE 2015 - Budapest, Hungary
Duration: 2015 Sep 92015 Sep 12

Publication series

NameStudies in Health Technology and Informatics
Volume217
ISSN (Print)0926-9630
ISSN (Electronic)1879-8365

Other

Other13th European Conference on the Advancement of Assistive Technology, AAATE 2015
CountryHungary
CityBudapest
Period15/9/915/9/12

Keywords

  • reading assistant
  • scene text recognition
  • text-to-speech
  • versatile text extraction
  • wearable camera

ASJC Scopus subject areas

  • Biomedical Engineering
  • Health Informatics
  • Health Information Management

Fingerprint Dive into the research topics of 'Versatile Text Extraction System for Text-to-Speech Reading Assistant Camera'. Together they form a unique fingerprint.

Cite this