Optical Text Recognition in Nepali and Bengali: A Transformer-based Approach (2404.02375v1)
Abstract: Efforts on the research and development of OCR systems for Low-Resource Languages are relatively new. Low-resource languages have little training data available for training Machine Translation systems or other systems. Even though a vast amount of text has been digitized and made available on the internet the text is still in PDF and Image format, which are not instantly accessible. This paper discusses text recognition for two scripts: Bengali and Nepali; there are about 300 and 40 million Bengali and Nepali speakers respectively. In this study, using encoder-decoder transformers, a model was developed, and its efficacy was assessed using a collection of optical text images, both handwritten and printed. The results signify that the suggested technique corresponds with current approaches and achieves high precision in recognizing text in Bengali and Nepali. This study can pave the way for the advanced and accessible study of linguistics in South East Asia.
- D. Paul and B. B. Chaudhuri, “A BLSTM Network for Printed Bengali OCR System with High Accuracy,” CoRR, vol. abs/1908.08674, 2019, [Online]. Available: http://arxiv.org/abs/1908.08674
- A. Sayeed, J. Shin, Md. A. M. Hasan, A. Y. Srizon, and M. Hasan, “BengaliNet: A Low-Cost Novel Convolutional Neural Network for Bengali Handwritten Characters Recognition,” Applied Sciences, vol. 11, no. 15, p. 6845, Jul. 2021, doi: 10.3390/app11156845.
- N. Das, S. Basu, R. Sarkar, M. Kundu, M. Nasipuri, and D. Basu, “An Improved Feature Descriptor for Recognition of Handwritten Bangla Alphabet,” arXiv (Cornell University), Jan. 2015, doi: 10.48550/arxiv.1501.05497.
- O. Ignat, “OCR Improves Machine Translation for Low-Resource Languages,” arXiv Cornell University), Feb. 27, 2022. https://arxiv.org/abs/2202.13274
- Mridha, Dr. M. F.; Quwsar Ohi, Abu; Ali, M. Ameer; Emon, Mazedul Islam; Kabir, Md Mohsin (2020), “BanglaWriting: A multi-purpose offline Bangla handwriting dataset”, Mendeley Data, V1, doi: 10.17632/r43wkvdk4w.1
- S M Rakib Hasan (3 papers)
- Aakar Dhakal (3 papers)
- Md Humaion Kabir Mehedi (3 papers)
- Annajiat Alim Rasel (8 papers)