A Hybrid Transformer–Classical Deep Learning OCR Framework for Multilingual Handwritten and Printed Document Recognition

View at publisher Summarise with VIRAT
Authors: D Mohammed Zilani, Mohebbanaaz
Cited by: 0
References: 17

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Sensible identification of printed and handwritten documents in multilingual is not an easy task, particularly to the languages with complicated scripts like Hindi, Telugu, Kannada, Bengali and Urdu. Conventional OCR engines have been found to show mixed results because of font variation, noisy input image as well as the structural nature of Indic scripts. This paper overcomes these drawbacks by developing a hybrid OCR model that incorporates three complementary recognition engines, namely, TrOCR (Transformer-based OCR of Vision), Tesseract (classical OCR), and EasyOCR (CNNLSTM OCR), in a common decision-making framework. In English text, the system suggested will carry out parallel OCR with TrOCR and Tesseract and then pick the most optimal output based on the score of confidence. In the case of Indian languages, EasyOCR has been used because it has a good recognition ability of Indic scripts and handwritten texts. Experimental results using six languages show that the hybrid system is always more effective than individual OCR engines with a maximum character accuracy of 98 percent on various printed and handwriting samples. The findings point to the usefulness of the synthesis of transformer based, classical and deep learning OCR to obtain stable multilingual document recognition applicable in the real-world digitization and automation process.

Fields of Study & Subjects

Handwritten Text Recognition Techniques Vehicle License Plate Recognition Advanced Neural Network Applications

Publication details

Journal / source
International Conference on Signal Processing and Electronics Design (ICSPED)
Pages
222–228
Publication date
23 January 2026
Language
EN