Development of an Intelligent OCR Model for Arabic Handwritten Text via Fine-Tuning of a Pretrained TrOCR Model

This work presents the development of an intelligent Optical Character Recognition (OCR) system tailored for Arabic handwritten text. Due to the cursive nature of Arabic script, variations in character shapes, and diverse handwriting styles, accurate recognition remains a significant challenge in computer vision and natural language processing. The project focuses on fine-tuning a pretrained Transformer-based OCR model (TrOCR) to improve recognition performance on Arabic handwriting. The approach aims to minimize Character Error Rate (CER) and Word Error Rate (WER) through model adaptation and data-driven optimization. The system integrates modern deep learning frameworks, including Python-based tools such as TensorFlow/PyTorch and OpenCV for preprocessing. Multiple datasets, including IFN/ENIT, AHCD, and synthetically generated handwritten samples, are utilized to enhance model robustness and generalization. Baseline comparisons are conducted using Tesseract OCR, alongside alternative architectures such as CRNN and attention-based sequence models. The document outlines the full pipeline, including preprocessing, model training, evaluation, and system design, supported by workflow diagrams and use-case illustrations. The results demonstrate the effectiveness of Transformer-based approaches for complex handwritten Arabic text recognition tasks. This work contributes to advancing OCR technologies for low-resource and complex scripts, with potential applications in digitization of historical documents, automated form processing, and Arabic language technologies. This work was conducted at Arab International University (AIU), Syria. The official website of the university is: https://www.aiu.edu.sy