Design of Sanskrit Devanagari Script Recognition and Sanskrit to English Translation System using deep learning.

Author

Dr. Aniruddha Kailuke, Rutuja Deshmukh, Bhavesh Bargat, Shital Tijare, Akshad Kurve

Abstract

 Sanskrit literature contains valuable knowledge in the fields of philosophy, linguistics, science, mathematics, medicine, and history. However, many Sanskrit documents are still preserved only as printed books, scanned pages, or handwritten manuscripts, making them difficult to search, edit, and translate. This review paper studies deep learning methods for designing a Sanskrit Devanagari script recognition and Sanskrit-to-English translation system. The objective of the system is to accept a Devanagari document image, recognize the Sanskrit text in editable Unicode form, and translate it into meaningful English prose. The reviewed methodology follows a modular pipeline consisting of image preprocessing, OCR recognition, text post-processing, tokenization, and neural machine translation. OpenCV-based preprocessing techniques such as grayscale conversion, Gaussian blur, Otsu thresholding, and morphological operations improve document clarity. For OCR, CNN/ResNet feature extraction, Bi-LSTM sequence modeling, and CTC decoding are suitable because they avoid character-level segmentation problems caused by shirorekha, matras, and conjunct characters. For translation, Transformer-based models such as IndicBART, mBART, or mT5 can be fine-tuned using Sanskrit-English parallel corpora and SentencePiece tokenization to handle Sandhi, Vibhakti, and flexible word order. The review finds that a two-stage OCR-plus-NMT architecture is practical, maintainable, and suitable for project implementation. Future work should focus on larger Sanskrit OCR datasets, improved manuscript restoration, grammar-aware tokenization, and human-in-the-loop correction.

Keywords

Sanskrit OCR; Devanagari script recognition; CNN-BiLSTM-CTC; IndicBART; neural machine translation; SentencePiece; deep learning.

Full Text:

Download Paper PDF

References

1.X. Zhang, Y. Wang, and L. Mishra, "Design of a Sanskrit Devanagari Script Recognition and Sanskrit Translation System Based on Deep Learning," IEEE Xplore Digital Library, vol. 24, no. 2, pp. 114–121, Feb. 2025.
2.R. Sharma, K. Purohit, and S. Joshi, "A Systematic Framework for Sanskrit Character Recognition Using Deep Learning," Electronic Letters on Computer Vision and Image Analysis (ELCVIA), vol. 24, no. 1, pp. 81–94, Jan. 2025.

3.A. Kumar and P. Dave, "Sanskrit to English Translation: A Comprehensive Survey and Implementation using Transformer Based Model," International Journal of Computer Applications, vol. 185, no. 28, pp. 14–22, May 2024.
4.R. Jayadevan, S. R. Kolhe, P. M. Patil, and U. Pal, "Online Recognition of Devanagari Script: A Survey," IEEE Transactions on Systems, Man, and Cybernetics, Part C (Applications and Reviews), vol. 41, no. 6, pp. 782–796, Nov. 2011.
5.S. Dhankhar, M. K. Gupta, F. H. Memon, S. Bhatia, P. Dadheech, and A. Mashat, "Support Vector Machine Based Handwritten Hindi Character Recognition and Summarization," Computer Systems Science and Engineering, vol. 43, no. 1, pp. 397–412, May 2022.
6.D. Kakwani et al., "IndicBART: A Pre-trained Model for Indic Natural Language Generation," in Findings of the Association for Computational Linguistics: ACL 2022, Dublin, Ireland, May 2022, pp. 3125–3136.

Share your valuable work from Social Media Buttons