Multi-Lingual Optical Character Recognition System Using the Reinforcement Learning of Character Segmenter
In this article, we present a new multi-lingual Optical Character Recognition (OCR) system for scanned documents. In the case of Latin characters, current open source systems such as Tesseract provide very high accuracy. However, the accuracy of the multi-lingual documents, including Asian character...
| Main Authors: | , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2020-01-01
|
| Series: | IEEE Access |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/9203882/ |
| _version_ | 1818873537458339840 |
|---|---|
| author | Jaewoo Park Eunji Lee Yoonsik Kim Isaac Kang Hyung Il Koo Nam Ik Cho |
| author_facet | Jaewoo Park Eunji Lee Yoonsik Kim Isaac Kang Hyung Il Koo Nam Ik Cho |
| author_sort | Jaewoo Park |
| collection | DOAJ |
| description | In this article, we present a new multi-lingual Optical Character Recognition (OCR) system for scanned documents. In the case of Latin characters, current open source systems such as Tesseract provide very high accuracy. However, the accuracy of the multi-lingual documents, including Asian characters, is usually lower than that for Latin-only documents. For example, when the document is the mix of English, Chinese and/or Korean characters, the OCR accuracy is lowered than English-only because the character/text properties of Chinese and Korean are quite different from Latin-type characters. To tackle these problems, we propose a new framework using three neural blocks (a segmenter, a switcher, and multiple recognizers) and the reinforcement learning of the segmenter: The segmenter partitions a given word image into multiple character images, the switcher assigns a recognizer for each sub-image, and the recognizers perform the recognition of assigned sub-images. The training of recognizers and switcher can be considered traditional image classification tasks and we train them with a supervised learning method. However, the supervised learning of the segmenter has two critical drawbacks: Its objective function is sub-optimal and its training requires a large amount of annotation efforts. Thus, by adopting the REINFORCE algorithm, we train the segmenter so as to optimize the overall performance, i.e., we minimize the edit distance of final recognition results. Experimental results have shown that the proposed method significantly improves the performance for multi-lingual scripts and large character set languages without using character boundary labels. |
| first_indexed | 2024-12-19T12:56:17Z |
| format | Article |
| id | doaj.art-e6acd7633bd847f2b5a4f89032f161ec |
| institution | Directory Open Access Journal |
| issn | 2169-3536 |
| language | English |
| last_indexed | 2024-12-19T12:56:17Z |
| publishDate | 2020-01-01 |
| publisher | IEEE |
| record_format | Article |
| series | IEEE Access |
| spelling | doaj.art-e6acd7633bd847f2b5a4f89032f161ec2022-12-21T20:20:22ZengIEEEIEEE Access2169-35362020-01-01817443717444810.1109/ACCESS.2020.30257699203882Multi-Lingual Optical Character Recognition System Using the Reinforcement Learning of Character SegmenterJaewoo Park0https://orcid.org/0000-0002-6816-4381Eunji Lee1https://orcid.org/0000-0002-7991-0618Yoonsik Kim2https://orcid.org/0000-0001-8023-8278Isaac Kang3Hyung Il Koo4https://orcid.org/0000-0002-6955-8083Nam Ik Cho5https://orcid.org/0000-0001-5297-4649Department of Electrical and Computer Engineering, INMC, Seoul National University, Seoul, South KoreaDepartment of Electrical and Computer Engineering, INMC, Seoul National University, Seoul, South KoreaDepartment of Electrical and Computer Engineering, INMC, Seoul National University, Seoul, South KoreaDepartment of Electrical and Computer Engineering, INMC, Seoul National University, Seoul, South KoreaDepartment of Electrical and Computer Engineering, Ajou University, Suwon, South KoreaDepartment of Electrical and Computer Engineering, INMC, Seoul National University, Seoul, South KoreaIn this article, we present a new multi-lingual Optical Character Recognition (OCR) system for scanned documents. In the case of Latin characters, current open source systems such as Tesseract provide very high accuracy. However, the accuracy of the multi-lingual documents, including Asian characters, is usually lower than that for Latin-only documents. For example, when the document is the mix of English, Chinese and/or Korean characters, the OCR accuracy is lowered than English-only because the character/text properties of Chinese and Korean are quite different from Latin-type characters. To tackle these problems, we propose a new framework using three neural blocks (a segmenter, a switcher, and multiple recognizers) and the reinforcement learning of the segmenter: The segmenter partitions a given word image into multiple character images, the switcher assigns a recognizer for each sub-image, and the recognizers perform the recognition of assigned sub-images. The training of recognizers and switcher can be considered traditional image classification tasks and we train them with a supervised learning method. However, the supervised learning of the segmenter has two critical drawbacks: Its objective function is sub-optimal and its training requires a large amount of annotation efforts. Thus, by adopting the REINFORCE algorithm, we train the segmenter so as to optimize the overall performance, i.e., we minimize the edit distance of final recognition results. Experimental results have shown that the proposed method significantly improves the performance for multi-lingual scripts and large character set languages without using character boundary labels.https://ieeexplore.ieee.org/document/9203882/Deep learningdocument analysisoptical character recognition |
| spellingShingle | Jaewoo Park Eunji Lee Yoonsik Kim Isaac Kang Hyung Il Koo Nam Ik Cho Multi-Lingual Optical Character Recognition System Using the Reinforcement Learning of Character Segmenter IEEE Access Deep learning document analysis optical character recognition |
| title | Multi-Lingual Optical Character Recognition System Using the Reinforcement Learning of Character Segmenter |
| title_full | Multi-Lingual Optical Character Recognition System Using the Reinforcement Learning of Character Segmenter |
| title_fullStr | Multi-Lingual Optical Character Recognition System Using the Reinforcement Learning of Character Segmenter |
| title_full_unstemmed | Multi-Lingual Optical Character Recognition System Using the Reinforcement Learning of Character Segmenter |
| title_short | Multi-Lingual Optical Character Recognition System Using the Reinforcement Learning of Character Segmenter |
| title_sort | multi lingual optical character recognition system using the reinforcement learning of character segmenter |
| topic | Deep learning document analysis optical character recognition |
| url | https://ieeexplore.ieee.org/document/9203882/ |
| work_keys_str_mv | AT jaewoopark multilingualopticalcharacterrecognitionsystemusingthereinforcementlearningofcharactersegmenter AT eunjilee multilingualopticalcharacterrecognitionsystemusingthereinforcementlearningofcharactersegmenter AT yoonsikkim multilingualopticalcharacterrecognitionsystemusingthereinforcementlearningofcharactersegmenter AT isaackang multilingualopticalcharacterrecognitionsystemusingthereinforcementlearningofcharactersegmenter AT hyungilkoo multilingualopticalcharacterrecognitionsystemusingthereinforcementlearningofcharactersegmenter AT namikcho multilingualopticalcharacterrecognitionsystemusingthereinforcementlearningofcharactersegmenter |