An Overview of Text Information Extraction from Images

An Overview of Text Information Extraction from Images

Prabhakar C. J. (Kuvempu University, India)
Copyright: © 2017 |Pages: 29
DOI: 10.4018/978-1-5225-2053-5.ch002
OnDemand PDF Download:
List Price: $37.50


In this chapter, we present an overview of text information extraction from images/video. This chapter starts with an introduction to computer vision and its applications, which is followed by an introduction to text information extraction from images/video. We describe various forms of text, challenges and steps involved in text information extraction process. The literature review of techniques for text information extraction from images/video is presented. Finally, our approach for extraction of scene text information from images is presented.
Chapter Preview


Digital images and videos form a larger part of archived multimedia data files, and they are rich in text information. Text present in images and videos provide valuable and important semantic information that may be of a particular interest as they are useful for describing the contents of an image. Text is therefore, becomes a Region of Interest (RoI), where, the points of interest must be clustered and extracted from the given image. Text Information Extraction (TIE) is concerned with the task of extracting relevant text information from digital images and videos. TIE system generally receives image or sequence of video frames as an input which can be either gray-scale or colored, compressed or un-compressed with still or moving text.

The TIE is a challenging problem because of variations occurring in images such as low contrast/resolutions, shaded or textured background, complex background, and variations in font size, style, artistic fonts, color, arbitrary text layouts, multi-script, effect of uncontrolled illumination, reflections, shadows, and distortion due to perspective projection (Crandall et al., 2003). These variations make the problem of automatic TIE extremely difficult. Figure 1 shows images containing variations in different aspects.

Figure 1.

Natural scene text images: images with variations in size, alignment, color, blur, illumination and distortion

Courtesy: Lee & Kim, 2013.

Applications of Text Information Extraction

  • License/Container Plate Extraction: Text information extraction techniques have been applied to automatically extract and recognize vehicle license plate and container plate, which helps for traffic surveillance and cargo container verification system (see Figure 2).

  • Address Block Extraction: Letter cover has a high degree of global structure among a limited number of entities. Mails usually have a printed label containing a block of address and are always pasted arbitrarily in any position. TIE can be used to extract the address block from mails.

  • Search for Web-Pages: Search engines have evolved over time due to growth of the web. Advanced search problems such as entity search and structured search on the web are facilitated by text information extraction.

  • Search for News Articles: Standard information extractions such as entity recognition help users to find specific pieces of information from news articles.

  • Page Segmentation: Document image analysis depends on the output of page segmentation technique which determines the format of a document page.

  • Retrieval of Intelligence Information: Vital information about suspicious criminal elements can be identified and retrieved from documents using TIE extraction techniques.

  • Biomedical: Discoveries related to biomedical field are searched and extracted from large scientific publications based on the identification of mentions on biomedical entities of interest from text and are linked to their corresponding entries in existing knowledge bases.

  • Navigation for Robots and Visually Impaired People: Navigation of autonomous mobile robots and navigation aid to visually impaired persons are facilitated by extraction and recognition of text information.

  • Automobile and Tourist Aid: Text information extraction and recognition are helpful for driving of automobiles and tourist systems.

Figure 2.

Sample images containing text information for various applications: (a),(b) license plate extraction/recognition, (c) container plate recognition and (d) address block localization

Courtesy of Matas & Zimmermann, 2005; ALPR-INFOSEC Institute; Adaptive Recognition-Hungary; Gaceb, Eglin, Lebourgeois & Emptoz,2009, respectively.

Complete Chapter List

Search this Book: