Efficient language-independent retrieval of printed documents without OCR

Walid Magdy, Kareem Darwish, Motaz El-Saban

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

1 Citation (Scopus)

Abstract

Recent book digitization initiatives have facilitated the access and search of millions of books. Although OCR remains essential for retrieving printed documents, OCR engines remain limited in the languages they handle and are generally expensive to build. This paper proposes a language independent approach that enables search through printed documents in a way that combines image-based matching with conventional IR techniques without using OCR. While image-based matching can be effective in finding similar words, complementing it with efficient retrieval techniques allows for sub-word matching, term weighting, and document ranking. The basic idea is that similar connected elements in printed documents are clustered and represented with ID's, which are then used to generate equivalent textual representations. The resultant representations are indexed using an IR engine and searched using the equivalent ID's of the connected elements in queries. Though, the main benefit of the proposed approach lies in languages for which no OCR exists, the technique was tested on English and Arabic to ascertain the relative effectiveness of the approach. The approach achieves more than 61% relative effectiveness compared to using OCR for both languages. While the reported numbers are lower than that of OCR-based approaches, the proposed method is fully automated, does not require any supervised training, and allows documents to be searchable within a few hours.

Original languageEnglish
Title of host publicationLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Pages334-343
Number of pages10
Volume5721 LNCS
DOIs
Publication statusPublished - 9 Nov 2009
Externally publishedYes
Event16th International Symposium on String Processing and Information Retrieval, SPIRE 2009 - Saariselka, Finland
Duration: 25 Aug 200927 Aug 2009

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume5721 LNCS
ISSN (Print)03029743
ISSN (Electronic)16113349

Other

Other16th International Symposium on String Processing and Information Retrieval, SPIRE 2009
CountryFinland
CitySaariselka
Period25/8/0927/8/09

Fingerprint

Optical character recognition
Retrieval
Engine
Subword
Digitization
Engines
Weighting
Ranking
Analog to digital conversion
Query
Language
Term

Keywords

  • Image based retrieval
  • OCR
  • Printed documents retrieval

ASJC Scopus subject areas

  • Computer Science(all)
  • Theoretical Computer Science

Cite this

Magdy, W., Darwish, K., & El-Saban, M. (2009). Efficient language-independent retrieval of printed documents without OCR. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 5721 LNCS, pp. 334-343). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 5721 LNCS). https://doi.org/10.1007/978-3-642-03784-9_33

Efficient language-independent retrieval of printed documents without OCR. / Magdy, Walid; Darwish, Kareem; El-Saban, Motaz.

Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). Vol. 5721 LNCS 2009. p. 334-343 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 5721 LNCS).

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

Magdy, W, Darwish, K & El-Saban, M 2009, Efficient language-independent retrieval of printed documents without OCR. in Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). vol. 5721 LNCS, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 5721 LNCS, pp. 334-343, 16th International Symposium on String Processing and Information Retrieval, SPIRE 2009, Saariselka, Finland, 25/8/09. https://doi.org/10.1007/978-3-642-03784-9_33
Magdy W, Darwish K, El-Saban M. Efficient language-independent retrieval of printed documents without OCR. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). Vol. 5721 LNCS. 2009. p. 334-343. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). https://doi.org/10.1007/978-3-642-03784-9_33
Magdy, Walid ; Darwish, Kareem ; El-Saban, Motaz. / Efficient language-independent retrieval of printed documents without OCR. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). Vol. 5721 LNCS 2009. pp. 334-343 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).
@inproceedings{c60246eec0194aabace626ad095909ab,
title = "Efficient language-independent retrieval of printed documents without OCR",
abstract = "Recent book digitization initiatives have facilitated the access and search of millions of books. Although OCR remains essential for retrieving printed documents, OCR engines remain limited in the languages they handle and are generally expensive to build. This paper proposes a language independent approach that enables search through printed documents in a way that combines image-based matching with conventional IR techniques without using OCR. While image-based matching can be effective in finding similar words, complementing it with efficient retrieval techniques allows for sub-word matching, term weighting, and document ranking. The basic idea is that similar connected elements in printed documents are clustered and represented with ID's, which are then used to generate equivalent textual representations. The resultant representations are indexed using an IR engine and searched using the equivalent ID's of the connected elements in queries. Though, the main benefit of the proposed approach lies in languages for which no OCR exists, the technique was tested on English and Arabic to ascertain the relative effectiveness of the approach. The approach achieves more than 61{\%} relative effectiveness compared to using OCR for both languages. While the reported numbers are lower than that of OCR-based approaches, the proposed method is fully automated, does not require any supervised training, and allows documents to be searchable within a few hours.",
keywords = "Image based retrieval, OCR, Printed documents retrieval",
author = "Walid Magdy and Kareem Darwish and Motaz El-Saban",
year = "2009",
month = "11",
day = "9",
doi = "10.1007/978-3-642-03784-9_33",
language = "English",
isbn = "3642037836",
volume = "5721 LNCS",
series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",
pages = "334--343",
booktitle = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",

}

TY - GEN

T1 - Efficient language-independent retrieval of printed documents without OCR

AU - Magdy, Walid

AU - Darwish, Kareem

AU - El-Saban, Motaz

PY - 2009/11/9

Y1 - 2009/11/9

N2 - Recent book digitization initiatives have facilitated the access and search of millions of books. Although OCR remains essential for retrieving printed documents, OCR engines remain limited in the languages they handle and are generally expensive to build. This paper proposes a language independent approach that enables search through printed documents in a way that combines image-based matching with conventional IR techniques without using OCR. While image-based matching can be effective in finding similar words, complementing it with efficient retrieval techniques allows for sub-word matching, term weighting, and document ranking. The basic idea is that similar connected elements in printed documents are clustered and represented with ID's, which are then used to generate equivalent textual representations. The resultant representations are indexed using an IR engine and searched using the equivalent ID's of the connected elements in queries. Though, the main benefit of the proposed approach lies in languages for which no OCR exists, the technique was tested on English and Arabic to ascertain the relative effectiveness of the approach. The approach achieves more than 61% relative effectiveness compared to using OCR for both languages. While the reported numbers are lower than that of OCR-based approaches, the proposed method is fully automated, does not require any supervised training, and allows documents to be searchable within a few hours.

AB - Recent book digitization initiatives have facilitated the access and search of millions of books. Although OCR remains essential for retrieving printed documents, OCR engines remain limited in the languages they handle and are generally expensive to build. This paper proposes a language independent approach that enables search through printed documents in a way that combines image-based matching with conventional IR techniques without using OCR. While image-based matching can be effective in finding similar words, complementing it with efficient retrieval techniques allows for sub-word matching, term weighting, and document ranking. The basic idea is that similar connected elements in printed documents are clustered and represented with ID's, which are then used to generate equivalent textual representations. The resultant representations are indexed using an IR engine and searched using the equivalent ID's of the connected elements in queries. Though, the main benefit of the proposed approach lies in languages for which no OCR exists, the technique was tested on English and Arabic to ascertain the relative effectiveness of the approach. The approach achieves more than 61% relative effectiveness compared to using OCR for both languages. While the reported numbers are lower than that of OCR-based approaches, the proposed method is fully automated, does not require any supervised training, and allows documents to be searchable within a few hours.

KW - Image based retrieval

KW - OCR

KW - Printed documents retrieval

UR - http://www.scopus.com/inward/record.url?scp=70350676786&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=70350676786&partnerID=8YFLogxK

U2 - 10.1007/978-3-642-03784-9_33

DO - 10.1007/978-3-642-03784-9_33

M3 - Conference contribution

SN - 3642037836

SN - 9783642037832

VL - 5721 LNCS

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 334

EP - 343

BT - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

ER -