Devanagari Text Recognition in Noisy Documents using Text Filters and Cooperative Deep Learning Methods

• Anup Kelkar,
• Parag Deshpande,
• O. G. Kakde,

Vol. 19, No. 11, November 30, 2025
10.3837/tiis.2025.11.001, Download Paper (Free):

• Camera-Captured Document Images
• Devanagari character’s geometric properties
• and OCR

Abstract

Optical Character Recognition (OCR) has become a crucial technology in text-based systems, facilitating the extraction of textual information from digitized images. While traditional OCR systems perform well on scanned documents, recognizing text from images captured by hand-held devices, particularly Devanagari scripts, presents unique challenges due to factors like variable lighting, complex backgrounds, and irregular font styles. This paper introduces a novel OCR system utilizing advanced image enhancement techniques and a filter-based convolutional neural network specifically designed to improve text recognition accuracy on Devanagari documents. Key contributions include innovative noise removal and the application of clustering techniques tailored to the distinctive features of the Devanagari script. Comparative analysis with existing OCR software demonstrates that the proposed system achieves superior accuracy, precision, recall, and F1-score, significantly reducing misclassification and error rates. The results suggest that the proposed system offers a robust solution for enhancing OCR performance in real-world applications involving complex Devanagari text recognition.

Statistics

Cite this article