Back to search

Democritus University of Thrace (DUTH)

Επεξεργασία ψηφιακών εγγραφών

Abstract

dc:description

This PhD dissertation presents new techniques for digital document processing. The goal of document processing is the extraction and the exploitation of the information that documents contain (text, images, graphical elements etc.). Document Image Processing (DIP) is a very active research field with many important applications. Optical Character Recognition (OCR), digital archives creation, the restoration of historical or other important documents, document image retrieval, compression and transmission, etc. are some examples of DIP applications. Four new DIP related techniques are presented. The first three proposed techniques aim to the extraction of text information from complex color documents such as cover books or magazine covers. The result of the first technique is used as input image from the second technique and the result of the second technique is used as input from the third technique. Although, these three techniques are applied on color documents, gray scale document images can also be used. The fourth proposed technique has been developed during the "POLYTIMO" research project and the goal is the segmentation of machine-printed historical documents. The first technique is a color segmentation procedure and the result can be used by any text information extraction method in order to locate the text areas or to extract the character components as a binary image result. The segmentation of the document is performed through a color reduction technique. The resulted number of colors is small so the document elements can be extracted as connected components. It is proposed that the initial image is pre-processed with an edge preserving smoothing filter in order to remove the noise, especially the texture type noise that results during the digitization process. Then, the image is sub-sampled with a new technique which has the ability to approximate the RGB color distribution. The last step utilizes the mean shift algorithm to locate the final color centers which are used to produce the final color reduction result. The result of the first technique is used by the second technique in order to locate the text areas, that is to border the areas in which homogeneous text exists. The method adopts the connected component based approach which is much faster compared with the texture based approach. For every color plane of the color reduced image, a binary image is created and on each one of these binary images, text is located. The final result is extracted by the superimposition of all text localization results. The proposed technique has the ability to locate horizontal and vertical text, even if these two text types exist in the same document image. The third technique is a color document binarization method. The text areas extracted from the second technique are used here and on each one, binarization techniques are applied. The resulted image consists of black pixel characters on white background. The advantage of processing independently the text areas is the use of local information instead of using global information. Local information of text areas provides more contrast between characters and background so these two elements can be separated more easily. Finally, the fourth technique is applied on binary document images which resulted from the digitization of historical books. The purpose is their segmentation at three levels, text lines, words and characters. The segmentation results can be used for document image retrieval and their transformation into editable electronic text through an OCR system. The main contribution here is that a modified Run Length Smoothing Algorithm (RLSA) is proposed, the ARLSA (Adaptive Run Length Smoothing Algorithm) in order to overcome its drawbacks when applied to this type of document images.

Degree

thesis:*
Grantor dc:publisher
Democritus University of Thrace (DUTH)
Year dc:date
2009

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Nikolaou, Nikolaos
  • Νικολάου, Νικόλαος

Subjects

dc:subject × 9

Rights

Language dc:language
gre

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/20924

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Nikolaou, Nikolaos; Νικολάου, Νικόλαος. Επεξεργασία ψηφιακών εγγραφών. Democritus University of Thrace (DUTH), 2009. http://hdl.handle.net/10442/hedi/20924