Cover Image

Unsupervised Document Binarization of Engineering Drawings via Multi Noise CycleGAN

The task of document binarization of degraded complex documents is tremendously challenging due to the various forms of noise often present in these documents. While the current state-of-the-art deep learning approaches are capable for the removal of various noise types in documents with high accura...

Full description

Saved in:
Bibliographic Details
Main Authors: Rosli, L.H., Hooi, Y.K., Bin, O.K.
Format: Article
Published: 2023
Online Access:http://scholars.utp.edu.my/id/eprint/37611/
https://www.scopus.com/inward/record.uri?eid=2-s2.0-85168803705&doi=10.14569%2fIJACSA.2023.0140791&partnerID=40&md5=af01ea6cb9d491b43f810d3933911448
Tags: Add Tag
No Tags, Be the first to tag this record!
id oai:scholars.utp.edu.my:37611
record_format eprints
spelling oai:scholars.utp.edu.my:376112023-10-13T13:04:59Z http://scholars.utp.edu.my/id/eprint/37611/ Unsupervised Document Binarization of Engineering Drawings via Multi Noise CycleGAN Rosli, L.H. Hooi, Y.K. Bin, O.K. The task of document binarization of degraded complex documents is tremendously challenging due to the various forms of noise often present in these documents. While the current state-of-the-art deep learning approaches are capable for the removal of various noise types in documents with high accuracy, they employ a supervised learning scheme which requires matching clean and noisy document image pairs which are difficult and costly to obtain for complex documents such as engineering drawings. In this paper, we propose our method for document binarization of engineering drawings using �Multi Noise CycleGAN�. The method utilizing unsupervised learning using adversarial and cycle-consistency loss is trained on unpaired noisy document images of various noise and image conditions. Experimental results for the removal of various noise types demonstrated that the method is able to reliably produce a clean image for any given noisy image and in certain noisy images achieve significant improvements over existing methods. © 2023, Science and Information Organization. All Rights Reserved. 2023 Article NonPeerReviewed Rosli, L.H. and Hooi, Y.K. and Bin, O.K. (2023) Unsupervised Document Binarization of Engineering Drawings via Multi Noise CycleGAN. International Journal of Advanced Computer Science and Applications, 14 (7). pp. 838-844. https://www.scopus.com/inward/record.uri?eid=2-s2.0-85168803705&doi=10.14569%2fIJACSA.2023.0140791&partnerID=40&md5=af01ea6cb9d491b43f810d3933911448 10.14569/IJACSA.2023.0140791 10.14569/IJACSA.2023.0140791 10.14569/IJACSA.2023.0140791
institution Universiti Teknologi Petronas
building UTP Resource Centre
collection Institutional Repository
continent Asia
country Malaysia
content_provider Universiti Teknologi Petronas
content_source UTP Institutional Repository
url_provider http://eprints.utp.edu.my/
description The task of document binarization of degraded complex documents is tremendously challenging due to the various forms of noise often present in these documents. While the current state-of-the-art deep learning approaches are capable for the removal of various noise types in documents with high accuracy, they employ a supervised learning scheme which requires matching clean and noisy document image pairs which are difficult and costly to obtain for complex documents such as engineering drawings. In this paper, we propose our method for document binarization of engineering drawings using �Multi Noise CycleGAN�. The method utilizing unsupervised learning using adversarial and cycle-consistency loss is trained on unpaired noisy document images of various noise and image conditions. Experimental results for the removal of various noise types demonstrated that the method is able to reliably produce a clean image for any given noisy image and in certain noisy images achieve significant improvements over existing methods. © 2023, Science and Information Organization. All Rights Reserved.
format Article
author Rosli, L.H.
Hooi, Y.K.
Bin, O.K.
spellingShingle Rosli, L.H.
Hooi, Y.K.
Bin, O.K.
Unsupervised Document Binarization of Engineering Drawings via Multi Noise CycleGAN
author_facet Rosli, L.H.
Hooi, Y.K.
Bin, O.K.
author_sort Rosli, L.H.
title Unsupervised Document Binarization of Engineering Drawings via Multi Noise CycleGAN
title_short Unsupervised Document Binarization of Engineering Drawings via Multi Noise CycleGAN
title_full Unsupervised Document Binarization of Engineering Drawings via Multi Noise CycleGAN
title_fullStr Unsupervised Document Binarization of Engineering Drawings via Multi Noise CycleGAN
title_full_unstemmed Unsupervised Document Binarization of Engineering Drawings via Multi Noise CycleGAN
title_sort unsupervised document binarization of engineering drawings via multi noise cyclegan
publishDate 2023
url http://scholars.utp.edu.my/id/eprint/37611/
https://www.scopus.com/inward/record.uri?eid=2-s2.0-85168803705&doi=10.14569%2fIJACSA.2023.0140791&partnerID=40&md5=af01ea6cb9d491b43f810d3933911448
_version_ 1781707932108324864
score 13.222552

Similar Items