Development of AI Technology for Automatic Translation of Korean Webtoon into Foreign Languages
0
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
Webtoon is a unique storytelling medium that has gained huge popularity in recent years. The increasing demands of the webtoons are mainly due to the uniqueness of Korean culture, and the availability of digital platforms. To address the global audience, these webtoons need to be translated into foreign languages. Existing human-based translation services are unable to meet the growing demands of the global audience because manual translation is often laborious and costly. Alternately, AI translation is expected to be much faster and more accurate. Text recognition is the first major step for webtoon translation. At present, Optical character recognition (OCR) techniques are only suitable for text detection inside the speech balloon, however, their performance drops on the onomatopoeic words. In this paper, we adopt a scene text recognition model for the training of onomatopoeic words. This model achieved 42% validation accuracy on our custom dataset which is fairly accurate for such challenging words. In the future, we plan to increase the dataset by adding more training examples, and through the database of onomatopoeia, we plan to build information on the comparison and usage of onomatopoeia in world languages.
Abstract
Webtoon is a unique storytelling medium that has gained huge popularity in recent years. The increasing demands of the webtoons are mainly due to the uniqueness of Korean culture, and the availability of digital platforms. To address the global audience, these webtoons need to be translated into foreign languages. Existing human-based translation services are unable to meet the growing demands of the global audience because manual translation is often laborious and costly. Alternately, AI translation is expected to be much faster and more accurate. Text recognition is the first major step for webtoon translation. At present, Optical character recognition (OCR) techniques are only suitable for text detection inside the speech balloon, however, their performance drops on the onomatopoeic words. In this paper, we adopt a scene text recognition model for the training of onomatopoeic words. This model achieved 42% validation accuracy on our custom dataset which is fairly accurate for such challenging words. In the future, we plan to increase the dataset by adding more training examples, and through the database of onomatopoeia, we plan to build information on the comparison and usage of onomatopoeia in world languages.
참고문헌 (11)
- Atienza, R., "International Conference on Document Analysis and Recognition", pp.319~334 (2021), Springer International Publishing
- Baek, J., "What is wrong with scene text recognition model comparisons? dataset and model analysis", Proceedings of the IEEE, 2019, 4715~4723
- Baek, J., "European Conference on Computer Vision", pp.267~283 (2022), Springer Nature Switzerland
- Hinami, R., "Towards fully automated manga translation", Proceedings of the AAAI Conference on Artificial Intelligence, 35(14), pp.12998~13008 (2021)
- Long, S., "Scene text detection and recognition: The deep learning era", International Journal of Computer Vision, 129, pp.161~184 (2021)
- Shi, B., "An end-to-end trainable neural network for image-based sequence recognition and its application to scene text recognition", IEEE transactions on pattern analysis and machine intelligence, 39(11), pp.2298~2304 (2016)
- Wu, L., "Editing text in the wild", Proceedings of the 27th ACM international conference on multimedia, pp.1500~1508 (2019)