Oblevit at AR-MS NAKBA NLP 2026 Subtask 2: Hybrid CNN–BiLSTM–CTC Framework with Linguistic Refinement for Arabic Handwritten Manuscript Recognition
Year: 2026
Type: conference-paper
Abstract: Arabic handwritten manuscript recognition remains a challenging problem in Optical Character Recognition (OCR) because of the cursive nature of the Arabic script, variations in handwriting styles, and character overlap.In addition, many Arabic letters are different only in the number or placement of dots, which increases ambiguity in recognition tasks.These challenges require models capable of extracting meaningful visual features while modeling long-range sequential dependencies.In this work, we propose a deep learning-based OCR system for Arabic handwritten text recognition.The proposed approach combines convolutional neural networks (CNN) for feature extraction, bidirectional recurrent neural networks (BiLSTM) for sequence modeling, and Connectionist Temporal Classification (CTC) for end-to-end training without character-level segmentation.To further enhance performance, decoding and specially designed post-processing strategies for Arabic text are applied to handle spacing issues and reduce character-level ambiguity.Experimental results show that the proposed method achieves competitive performance and effectively handles complex handwritten Arabic manuscripts.
Institution: Arab Open University
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7166529330