A Deep Transfer Learning Approach to Document Image Quality Assessment

Tan Lu; Ann Dooms

doi:10.1109/ICDAR.2019.00221

A Deep Transfer Learning Approach to Document Image Quality Assessment

Tan Lu, Ann Dooms

Research output: Chapter in Book/Report/Conference proceeding › Conference paper

13 Citations (Scopus)

Abstract

Document image quality assessment (DIQA) is an important process for various applications such as optical character recognition (OCR) and document restoration. In this paper we propose a no-reference DIQA model based on a deep convolutional neural network (DCNN), where the rich knowledge of natural scene image characterization of a previously-trained DCNN is exploited towards OCR accuracy oriented document image quality assessment. Following a two-stage deep transfer learning procedure, we fine-tune the knowledge base of the DCNN in the first phase and bring in a task-specific segment consisting of three fully connected (FC) layers in the second phase. Based on the fine-tuned knowledge base, the task-specific segment is trained from scratch to facilitate the application of the transferred knowledge on the new task of document quality assessment. Testing results on a benchmark dataset demonstrate that the proposed model achieves state-of-the-art performance.

Original language	English
Title of host publication	2019 International Conference on Document Analysis and Recognition (ICDAR)
Publisher	IEEE
Pages	1372-1377
Number of pages	6
ISBN (Print)	978-1-7281-3014-9, 978-1-7281-3015-6
DOIs	https://doi.org/10.1109/ICDAR.2019.00221
Publication status	Published - 3 Feb 2020
Event	15th International Conference on Document Analysis and Recognition - International Convention Centre Sydney, Sydney, Australia Duration: 20 Sept 2019 → 25 Sept 2019 https://icdar2019.org/

Publication series

Name	Proceedings of the International Conference on Document Analysis and Recognition, ICDAR
ISSN (Print)	1520-5363

Conference

Conference	15th International Conference on Document Analysis and Recognition
Abbreviated title	ICDAR2019
Country/Territory	Australia
City	Sydney
Period	20/09/19 → 25/09/19
Internet address	https://icdar2019.org/

Bibliographical note

Funding Information:
ACKNOWLEDGMENT This research is supported by the Auditing Digitisation Outputs in the Cultural Heritage Sector (ADOCHS) project (Contract No. BR/154/A6/ADOCHS), financed by the Belgian Science Policy (Belspo) within the scope of the BRAIN programme.

Publisher Copyright:
© 2019 IEEE.

Access to Document

10.1109/ICDAR.2019.00221

Cite this

@inproceedings{b0438802fed4423287130773a34e2521,

title = "A Deep Transfer Learning Approach to Document Image Quality Assessment",

abstract = "Document image quality assessment (DIQA) is an important process for various applications such as optical character recognition (OCR) and document restoration. In this paper we propose a no-reference DIQA model based on a deep convolutional neural network (DCNN), where the rich knowledge of natural scene image characterization of a previously-trained DCNN is exploited towards OCR accuracy oriented document image quality assessment. Following a two-stage deep transfer learning procedure, we fine-tune the knowledge base of the DCNN in the first phase and bring in a task-specific segment consisting of three fully connected (FC) layers in the second phase. Based on the fine-tuned knowledge base, the task-specific segment is trained from scratch to facilitate the application of the transferred knowledge on the new task of document quality assessment. Testing results on a benchmark dataset demonstrate that the proposed model achieves state-of-the-art performance.",

author = "Tan Lu and Ann Dooms",

note = "Funding Information: ACKNOWLEDGMENT This research is supported by the Auditing Digitisation Outputs in the Cultural Heritage Sector (ADOCHS) project (Contract No. BR/154/A6/ADOCHS), financed by the Belgian Science Policy (Belspo) within the scope of the BRAIN programme. Publisher Copyright: {\textcopyright} 2019 IEEE.; 15th International Conference on Document Analysis and Recognition ; Conference date: 20-09-2019 Through 25-09-2019",

year = "2020",

month = feb,

day = "3",

doi = "10.1109/ICDAR.2019.00221",

language = "English",

isbn = "978-1-7281-3014-9",

series = "Proceedings of the International Conference on Document Analysis and Recognition, ICDAR",

publisher = "IEEE",

pages = "1372--1377",

booktitle = "2019 International Conference on Document Analysis and Recognition (ICDAR)",

url = "https://icdar2019.org/",

}

Lu, T & Dooms, A 2020, A Deep Transfer Learning Approach to Document Image Quality Assessment. in 2019 International Conference on Document Analysis and Recognition (ICDAR). Proceedings of the International Conference on Document Analysis and Recognition, ICDAR, IEEE, pp. 1372-1377, 15th International Conference on Document Analysis and Recognition, Sydney, Australia, 20/09/19. https://doi.org/10.1109/ICDAR.2019.00221

TY - GEN

T1 - A Deep Transfer Learning Approach to Document Image Quality Assessment

AU - Lu, Tan

AU - Dooms, Ann

N1 - Funding Information: ACKNOWLEDGMENT This research is supported by the Auditing Digitisation Outputs in the Cultural Heritage Sector (ADOCHS) project (Contract No. BR/154/A6/ADOCHS), financed by the Belgian Science Policy (Belspo) within the scope of the BRAIN programme. Publisher Copyright: © 2019 IEEE.

PY - 2020/2/3

Y1 - 2020/2/3

N2 - Document image quality assessment (DIQA) is an important process for various applications such as optical character recognition (OCR) and document restoration. In this paper we propose a no-reference DIQA model based on a deep convolutional neural network (DCNN), where the rich knowledge of natural scene image characterization of a previously-trained DCNN is exploited towards OCR accuracy oriented document image quality assessment. Following a two-stage deep transfer learning procedure, we fine-tune the knowledge base of the DCNN in the first phase and bring in a task-specific segment consisting of three fully connected (FC) layers in the second phase. Based on the fine-tuned knowledge base, the task-specific segment is trained from scratch to facilitate the application of the transferred knowledge on the new task of document quality assessment. Testing results on a benchmark dataset demonstrate that the proposed model achieves state-of-the-art performance.

AB - Document image quality assessment (DIQA) is an important process for various applications such as optical character recognition (OCR) and document restoration. In this paper we propose a no-reference DIQA model based on a deep convolutional neural network (DCNN), where the rich knowledge of natural scene image characterization of a previously-trained DCNN is exploited towards OCR accuracy oriented document image quality assessment. Following a two-stage deep transfer learning procedure, we fine-tune the knowledge base of the DCNN in the first phase and bring in a task-specific segment consisting of three fully connected (FC) layers in the second phase. Based on the fine-tuned knowledge base, the task-specific segment is trained from scratch to facilitate the application of the transferred knowledge on the new task of document quality assessment. Testing results on a benchmark dataset demonstrate that the proposed model achieves state-of-the-art performance.

UR - https://cris.vub.be/en/publications/a-deep-transfer-learning-approach-to-document-image-quality-assessment(cfa76a91-72b6-4442-9f6c-2d3d61eb2ba9).html

UR - http://www.scopus.com/inward/record.url?scp=85079862564&partnerID=8YFLogxK

U2 - 10.1109/ICDAR.2019.00221

DO - 10.1109/ICDAR.2019.00221

M3 - Conference paper

SN - 978-1-7281-3014-9

SN - 978-1-7281-3015-6

T3 - Proceedings of the International Conference on Document Analysis and Recognition, ICDAR

SP - 1372

EP - 1377

BT - 2019 International Conference on Document Analysis and Recognition (ICDAR)

PB - IEEE

T2 - 15th International Conference on Document Analysis and Recognition

Y2 - 20 September 2019 through 25 September 2019

ER -

A Deep Transfer Learning Approach to Document Image Quality Assessment

Abstract

Publication series

Conference

Bibliographical note

Access to Document

Other files and links

Fingerprint

Cite this