Amharic OCR: An End-to-End Learning
In this paper, we introduce an end-to-end Amharic text-line image recognition approach based on recurrent neural networks. Amharic is an indigenous Ethiopic script which follows a unique syllabic writing system adopted from an ancient Geez script. This script uses 34 consonant characters with the se...
Main Authors: | , , , , , |
---|---|
Format: | Article |
Language: | English |
Published: |
MDPI AG
2020-02-01
|
Series: | Applied Sciences |
Subjects: | |
Online Access: | https://www.mdpi.com/2076-3417/10/3/1117 |
id |
doaj-f9807535c08a468586a6979c4612533f |
---|---|
record_format |
Article |
spelling |
doaj-f9807535c08a468586a6979c4612533f2020-11-25T00:19:32ZengMDPI AGApplied Sciences2076-34172020-02-01103111710.3390/app10031117app10031117Amharic OCR: An End-to-End LearningBirhanu Belay0Tewodros Habtegebrial1Million Meshesha2Marcus Liwicki3Gebeyehu Belay4Didier Stricker5Department of Computer Science, University of Kaiserslautern, 67653 Kaiserslautern, GermanyDepartment of Computer Science, University of Kaiserslautern, 67653 Kaiserslautern, GermanySchool of Information Science, Addis Ababa University, Addis Ababa, EthiopiaDepartment of Computer Science, Lulea University of Technology, 97187 Lulea, SwedenFaculty of Computing, Bahir Dar Institute of Technology, Bahir Dar, EthiopiaDepartment of Computer Science, University of Kaiserslautern, 67653 Kaiserslautern, GermanyIn this paper, we introduce an end-to-end Amharic text-line image recognition approach based on recurrent neural networks. Amharic is an indigenous Ethiopic script which follows a unique syllabic writing system adopted from an ancient Geez script. This script uses 34 consonant characters with the seven vowel variants of each (called basic characters) and other labialized characters derived by adding diacritical marks and/or removing parts of the basic characters. These associated diacritics on basic characters are relatively smaller in size, visually similar, and challenging to distinguish from the derived characters. Motivated by the recent success of end-to-end learning in pattern recognition, we propose a model which integrates a feature extractor, sequence learner, and transcriber in a unified module and then trained in an end-to-end fashion. The experimental results, on a printed and synthetic benchmark Amharic Optical Character Recognition (OCR) database called ADOCR, demonstrated that the proposed model outperforms state-of-the-art methods by 6.98% and 1.05%, respectively.https://www.mdpi.com/2076-3417/10/3/1117amharic scriptcnnctcend-to-end learninglstmocrpattern recognitiontext-line image |
collection |
DOAJ |
language |
English |
format |
Article |
sources |
DOAJ |
author |
Birhanu Belay Tewodros Habtegebrial Million Meshesha Marcus Liwicki Gebeyehu Belay Didier Stricker |
spellingShingle |
Birhanu Belay Tewodros Habtegebrial Million Meshesha Marcus Liwicki Gebeyehu Belay Didier Stricker Amharic OCR: An End-to-End Learning Applied Sciences amharic script cnn ctc end-to-end learning lstm ocr pattern recognition text-line image |
author_facet |
Birhanu Belay Tewodros Habtegebrial Million Meshesha Marcus Liwicki Gebeyehu Belay Didier Stricker |
author_sort |
Birhanu Belay |
title |
Amharic OCR: An End-to-End Learning |
title_short |
Amharic OCR: An End-to-End Learning |
title_full |
Amharic OCR: An End-to-End Learning |
title_fullStr |
Amharic OCR: An End-to-End Learning |
title_full_unstemmed |
Amharic OCR: An End-to-End Learning |
title_sort |
amharic ocr: an end-to-end learning |
publisher |
MDPI AG |
series |
Applied Sciences |
issn |
2076-3417 |
publishDate |
2020-02-01 |
description |
In this paper, we introduce an end-to-end Amharic text-line image recognition approach based on recurrent neural networks. Amharic is an indigenous Ethiopic script which follows a unique syllabic writing system adopted from an ancient Geez script. This script uses 34 consonant characters with the seven vowel variants of each (called basic characters) and other labialized characters derived by adding diacritical marks and/or removing parts of the basic characters. These associated diacritics on basic characters are relatively smaller in size, visually similar, and challenging to distinguish from the derived characters. Motivated by the recent success of end-to-end learning in pattern recognition, we propose a model which integrates a feature extractor, sequence learner, and transcriber in a unified module and then trained in an end-to-end fashion. The experimental results, on a printed and synthetic benchmark Amharic Optical Character Recognition (OCR) database called ADOCR, demonstrated that the proposed model outperforms state-of-the-art methods by 6.98% and 1.05%, respectively. |
topic |
amharic script cnn ctc end-to-end learning lstm ocr pattern recognition text-line image |
url |
https://www.mdpi.com/2076-3417/10/3/1117 |
work_keys_str_mv |
AT birhanubelay amharicocranendtoendlearning AT tewodroshabtegebrial amharicocranendtoendlearning AT millionmeshesha amharicocranendtoendlearning AT marcusliwicki amharicocranendtoendlearning AT gebeyehubelay amharicocranendtoendlearning AT didierstricker amharicocranendtoendlearning |
_version_ |
1725371518632853504 |