Improved BLSTM neural networks for recognition of on-line bangla complex words

Volkmar Frinken, Nilanjana Bhattacharya, Seiichi Uchida, Umapada Pal

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

7 Citations (Scopus)

Abstract

While bi-directional long short-term (BLSTM) neural network have been demonstrated to perform very well for English or Arabic, the huge number of different output classes (characters) encountered in many Asian fonts, poses a severe challenge. In this work we investigate different encoding schemes of Bangla compound characters and compare the recognition accuracies. We propose to model complex characters not as unique symbols, which are represented by individual nodes in the output layer. Instead, we exploit the property of long-distance-dependent classification in BLSTM neural networks. We classify only basic strokes and use special nodes which react to semantic changes in the writing, i.e., distinguishing inter-character spaces from intra-character spaces. We show that our approach outperforms the common approaches to BLSTM neural network-based handwriting recognition considerably.

Original languageEnglish
Title of host publicationStructural, Syntactic, and Statistical Pattern Recognition - Joint IAPR International Workshop, S+SSPR 2014, Proceedings
PublisherSpringer Verlag
Pages404-413
Number of pages10
ISBN (Print)9783662444146
DOIs
Publication statusPublished - Jan 1 2014
EventJoint IAPR International Workshop on Structural, Syntactic, and Statistical Pattern Recognition, S+SSPR 2014 - Joensuu, Finland
Duration: Aug 20 2014Aug 22 2014

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume8621 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Other

OtherJoint IAPR International Workshop on Structural, Syntactic, and Statistical Pattern Recognition, S+SSPR 2014
CountryFinland
CityJoensuu
Period8/20/148/22/14

Fingerprint

Neural Networks
Neural networks
Handwriting Recognition
Semantics
Output
Vertex of a graph
Stroke
Encoding
Classify
Character
Dependent
Model

All Science Journal Classification (ASJC) codes

  • Theoretical Computer Science
  • Computer Science(all)

Cite this

Frinken, V., Bhattacharya, N., Uchida, S., & Pal, U. (2014). Improved BLSTM neural networks for recognition of on-line bangla complex words. In Structural, Syntactic, and Statistical Pattern Recognition - Joint IAPR International Workshop, S+SSPR 2014, Proceedings (pp. 404-413). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 8621 LNCS). Springer Verlag. https://doi.org/10.1007/978-3-662-44415-3_41

Improved BLSTM neural networks for recognition of on-line bangla complex words. / Frinken, Volkmar; Bhattacharya, Nilanjana; Uchida, Seiichi; Pal, Umapada.

Structural, Syntactic, and Statistical Pattern Recognition - Joint IAPR International Workshop, S+SSPR 2014, Proceedings. Springer Verlag, 2014. p. 404-413 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 8621 LNCS).

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

Frinken, V, Bhattacharya, N, Uchida, S & Pal, U 2014, Improved BLSTM neural networks for recognition of on-line bangla complex words. in Structural, Syntactic, and Statistical Pattern Recognition - Joint IAPR International Workshop, S+SSPR 2014, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 8621 LNCS, Springer Verlag, pp. 404-413, Joint IAPR International Workshop on Structural, Syntactic, and Statistical Pattern Recognition, S+SSPR 2014, Joensuu, Finland, 8/20/14. https://doi.org/10.1007/978-3-662-44415-3_41
Frinken V, Bhattacharya N, Uchida S, Pal U. Improved BLSTM neural networks for recognition of on-line bangla complex words. In Structural, Syntactic, and Statistical Pattern Recognition - Joint IAPR International Workshop, S+SSPR 2014, Proceedings. Springer Verlag. 2014. p. 404-413. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). https://doi.org/10.1007/978-3-662-44415-3_41
Frinken, Volkmar ; Bhattacharya, Nilanjana ; Uchida, Seiichi ; Pal, Umapada. / Improved BLSTM neural networks for recognition of on-line bangla complex words. Structural, Syntactic, and Statistical Pattern Recognition - Joint IAPR International Workshop, S+SSPR 2014, Proceedings. Springer Verlag, 2014. pp. 404-413 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).
@inproceedings{626f5662d0e047d49e2a2b1fd0bf39a9,
title = "Improved BLSTM neural networks for recognition of on-line bangla complex words",
abstract = "While bi-directional long short-term (BLSTM) neural network have been demonstrated to perform very well for English or Arabic, the huge number of different output classes (characters) encountered in many Asian fonts, poses a severe challenge. In this work we investigate different encoding schemes of Bangla compound characters and compare the recognition accuracies. We propose to model complex characters not as unique symbols, which are represented by individual nodes in the output layer. Instead, we exploit the property of long-distance-dependent classification in BLSTM neural networks. We classify only basic strokes and use special nodes which react to semantic changes in the writing, i.e., distinguishing inter-character spaces from intra-character spaces. We show that our approach outperforms the common approaches to BLSTM neural network-based handwriting recognition considerably.",
author = "Volkmar Frinken and Nilanjana Bhattacharya and Seiichi Uchida and Umapada Pal",
year = "2014",
month = "1",
day = "1",
doi = "10.1007/978-3-662-44415-3_41",
language = "English",
isbn = "9783662444146",
series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",
publisher = "Springer Verlag",
pages = "404--413",
booktitle = "Structural, Syntactic, and Statistical Pattern Recognition - Joint IAPR International Workshop, S+SSPR 2014, Proceedings",
address = "Germany",

}

TY - GEN

T1 - Improved BLSTM neural networks for recognition of on-line bangla complex words

AU - Frinken, Volkmar

AU - Bhattacharya, Nilanjana

AU - Uchida, Seiichi

AU - Pal, Umapada

PY - 2014/1/1

Y1 - 2014/1/1

N2 - While bi-directional long short-term (BLSTM) neural network have been demonstrated to perform very well for English or Arabic, the huge number of different output classes (characters) encountered in many Asian fonts, poses a severe challenge. In this work we investigate different encoding schemes of Bangla compound characters and compare the recognition accuracies. We propose to model complex characters not as unique symbols, which are represented by individual nodes in the output layer. Instead, we exploit the property of long-distance-dependent classification in BLSTM neural networks. We classify only basic strokes and use special nodes which react to semantic changes in the writing, i.e., distinguishing inter-character spaces from intra-character spaces. We show that our approach outperforms the common approaches to BLSTM neural network-based handwriting recognition considerably.

AB - While bi-directional long short-term (BLSTM) neural network have been demonstrated to perform very well for English or Arabic, the huge number of different output classes (characters) encountered in many Asian fonts, poses a severe challenge. In this work we investigate different encoding schemes of Bangla compound characters and compare the recognition accuracies. We propose to model complex characters not as unique symbols, which are represented by individual nodes in the output layer. Instead, we exploit the property of long-distance-dependent classification in BLSTM neural networks. We classify only basic strokes and use special nodes which react to semantic changes in the writing, i.e., distinguishing inter-character spaces from intra-character spaces. We show that our approach outperforms the common approaches to BLSTM neural network-based handwriting recognition considerably.

UR - http://www.scopus.com/inward/record.url?scp=84906309034&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84906309034&partnerID=8YFLogxK

U2 - 10.1007/978-3-662-44415-3_41

DO - 10.1007/978-3-662-44415-3_41

M3 - Conference contribution

AN - SCOPUS:84906309034

SN - 9783662444146

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

SP - 404

EP - 413

BT - Structural, Syntactic, and Statistical Pattern Recognition - Joint IAPR International Workshop, S+SSPR 2014, Proceedings

PB - Springer Verlag

ER -