Recognition of Anomalously Deformed Kana Sequences in Japanese Historical Documents

Nam Tuan LY; Kha Cong NGUYEN; Cuong Tuan NGUYEN; Masaki NAKAGAWA

doi:10.1587/transinf.2018EDP7361

IEICE TRANSACTIONS on Information

Recognition of Anomalously Deformed Kana Sequences in Japanese Historical Documents

Nam Tuan LY, Kha Cong NGUYEN, Cuong Tuan NGUYEN, Masaki NAKAGAWA

Full Text Views

0

Cite this

Summary :

This paper presents recognition of anomalously deformed Kana sequences in Japanese historical documents, for which a contest was held by IEICE PRMU 2017. The contest was divided into three levels in accordance with the number of characters to be recognized: level 1: single characters, level 2: sequences of three vertically written Kana characters, and level 3: unrestricted sets of characters composed of three or more characters possibly in multiple lines. This paper focuses on the methods for levels 2 and 3 that won the contest. We basically follow the segmentation-free approach and employ the hierarchy of a Convolutional Neural Network (CNN) for feature extraction, Bidirectional Long Short-Term Memory (BLSTM) for frame prediction, and Connectionist Temporal Classification (CTC) for text recognition, which is named a Deep Convolutional Recurrent Network (DCRN). We compare the pretrained CNN approach and the end-to-end approach with more detailed variations for level 2. Then, we propose a method of vertical text line segmentation and multiple line concatenation before applying DCRN for level 3. We also examine a two-dimensional BLSTM (2DBLSTM) based method for level 3. We present the evaluation of the best methods by cross validation. We achieved an accuracy of 89.10% for the three-Kana-character sequence recognition and an accuracy of 87.70% for the unrestricted Kana recognition without employing linguistic context. These results prove the performances of the proposed models on the level 2 and 3 tasks.

Publication: IEICE TRANSACTIONS on Information Vol.E102-D No.8 pp.1554-1564

Publication Date: 2019/08/01

Publicized: 2019/05/07

Online ISSN: 1745-1361

DOI: 10.1587/transinf.2018EDP7361

Type of Manuscript: PAPER

Category: Image Recognition, Computer Vision

Authors

Nam Tuan LY
  Tokyo University of Agriculture and Technology
Kha Cong NGUYEN
  Tokyo University of Agriculture and Technology
Cuong Tuan NGUYEN
  Tokyo University of Agriculture and Technology
Masaki NAKAGAWA
  Tokyo University of Agriculture and Technology

Keyword

historical documents, deformed kana recognition, handwriting recognition, deep neural networks

Cite this

Copy

Nam Tuan LY, Kha Cong NGUYEN, Cuong Tuan NGUYEN, Masaki NAKAGAWA, "Recognition of Anomalously Deformed Kana Sequences in Japanese Historical Documents" in IEICE TRANSACTIONS on Information, vol. E102-D, no. 8, pp. 1554-1564, August 2019, doi: 10.1587/transinf.2018EDP7361.
Abstract: This paper presents recognition of anomalously deformed Kana sequences in Japanese historical documents, for which a contest was held by IEICE PRMU 2017. The contest was divided into three levels in accordance with the number of characters to be recognized: level 1: single characters, level 2: sequences of three vertically written Kana characters, and level 3: unrestricted sets of characters composed of three or more characters possibly in multiple lines. This paper focuses on the methods for levels 2 and 3 that won the contest. We basically follow the segmentation-free approach and employ the hierarchy of a Convolutional Neural Network (CNN) for feature extraction, Bidirectional Long Short-Term Memory (BLSTM) for frame prediction, and Connectionist Temporal Classification (CTC) for text recognition, which is named a Deep Convolutional Recurrent Network (DCRN). We compare the pretrained CNN approach and the end-to-end approach with more detailed variations for level 2. Then, we propose a method of vertical text line segmentation and multiple line concatenation before applying DCRN for level 3. We also examine a two-dimensional BLSTM (2DBLSTM) based method for level 3. We present the evaluation of the best methods by cross validation. We achieved an accuracy of 89.10% for the three-Kana-character sequence recognition and an accuracy of 87.70% for the unrestricted Kana recognition without employing linguistic context. These results prove the performances of the proposed models on the level 2 and 3 tasks.
URL: https://global.ieice.org/en_transactions/information/10.1587/transinf.2018EDP7361/_p

Copy

@ARTICLE{e102-d_8_1554,
author={Nam Tuan LY, Kha Cong NGUYEN, Cuong Tuan NGUYEN, Masaki NAKAGAWA, },
journal={IEICE TRANSACTIONS on Information},
title={Recognition of Anomalously Deformed Kana Sequences in Japanese Historical Documents},
year={2019},
volume={E102-D},
number={8},
pages={1554-1564},
abstract={This paper presents recognition of anomalously deformed Kana sequences in Japanese historical documents, for which a contest was held by IEICE PRMU 2017. The contest was divided into three levels in accordance with the number of characters to be recognized: level 1: single characters, level 2: sequences of three vertically written Kana characters, and level 3: unrestricted sets of characters composed of three or more characters possibly in multiple lines. This paper focuses on the methods for levels 2 and 3 that won the contest. We basically follow the segmentation-free approach and employ the hierarchy of a Convolutional Neural Network (CNN) for feature extraction, Bidirectional Long Short-Term Memory (BLSTM) for frame prediction, and Connectionist Temporal Classification (CTC) for text recognition, which is named a Deep Convolutional Recurrent Network (DCRN). We compare the pretrained CNN approach and the end-to-end approach with more detailed variations for level 2. Then, we propose a method of vertical text line segmentation and multiple line concatenation before applying DCRN for level 3. We also examine a two-dimensional BLSTM (2DBLSTM) based method for level 3. We present the evaluation of the best methods by cross validation. We achieved an accuracy of 89.10% for the three-Kana-character sequence recognition and an accuracy of 87.70% for the unrestricted Kana recognition without employing linguistic context. These results prove the performances of the proposed models on the level 2 and 3 tasks.},
keywords={},
doi={10.1587/transinf.2018EDP7361},
ISSN={1745-1361},
month={August},}

Copy

TY - JOUR
TI - Recognition of Anomalously Deformed Kana Sequences in Japanese Historical Documents
T2 - IEICE TRANSACTIONS on Information
SP - 1554
EP - 1564
AU - Nam Tuan LY
AU - Kha Cong NGUYEN
AU - Cuong Tuan NGUYEN
AU - Masaki NAKAGAWA
PY - 2019
DO - 10.1587/transinf.2018EDP7361
JO - IEICE TRANSACTIONS on Information
SN - 1745-1361
VL - E102-D
IS - 8
JA - IEICE TRANSACTIONS on Information
Y1 - August 2019
AB - This paper presents recognition of anomalously deformed Kana sequences in Japanese historical documents, for which a contest was held by IEICE PRMU 2017. The contest was divided into three levels in accordance with the number of characters to be recognized: level 1: single characters, level 2: sequences of three vertically written Kana characters, and level 3: unrestricted sets of characters composed of three or more characters possibly in multiple lines. This paper focuses on the methods for levels 2 and 3 that won the contest. We basically follow the segmentation-free approach and employ the hierarchy of a Convolutional Neural Network (CNN) for feature extraction, Bidirectional Long Short-Term Memory (BLSTM) for frame prediction, and Connectionist Temporal Classification (CTC) for text recognition, which is named a Deep Convolutional Recurrent Network (DCRN). We compare the pretrained CNN approach and the end-to-end approach with more detailed variations for level 2. Then, we propose a method of vertical text line segmentation and multiple line concatenation before applying DCRN for level 3. We also examine a two-dimensional BLSTM (2DBLSTM) based method for level 3. We present the evaluation of the best methods by cross validation. We achieved an accuracy of 89.10% for the three-Kana-character sequence recognition and an accuracy of 87.70% for the unrestricted Kana recognition without employing linguistic context. These results prove the performances of the proposed models on the level 2 and 3 tasks.
ER -

IEICE TRANSACTIONS on Information