CNN models with attention reach 99.05% top-1 accuracy on line-level splits and 78.61% on page-disjoint splits for writer identification after expanding the labeled portion of the Muharaf historical Arabic manuscript dataset.
arXiv preprint arXiv:2410.02179 (2024)
3 Pith papers cite this work. Polarity classification is still indexing.
3
Pith papers citing it
fields
cs.CV 3years
2026 3representative citing papers
Controlled experiments across nine datasets show Arabic HTR maintains a 5-7 CER gap over Latin even at full data scale, with Arabic requiring more samples due to heavier-tailed character distributions and more visually similar confusions.
Joint training of CRNN and HTR-VT models across Arabic-script datasets yields lower character error rates than single-language training in low-resource regimes, reaching 9.99 CER on Persian and 14.45 on Urdu.
citing papers explorer
-
Performance Gap Analysis between Latin and Arabic Scripts HTR
Controlled experiments across nine datasets show Arabic HTR maintains a 5-7 CER gap over Latin even at full data scale, with Arabic requiring more samples due to heavier-tailed character distributions and more visually similar confusions.