An Intelligent OCR Approach for Arabic Handwriting Using a Fine-Tuned TrOCR Model

This report presents the development of an intelligent Optical Character Recognition (OCR) system designed specifically for Arabic handwritten text using a fine-tuned Transformer-based model (TrOCR). Arabic handwriting recognition is particularly challenging due to the cursive nature of the script, context-dependent letter forms, and high variability in individual writing styles. To address these challenges, the proposed system leverages a state-of-the-art vision-language architecture that combines a Vision Transformer (ViT) encoder with a Transformer-based decoder, fine-tuned on benchmark datasets such as IFN/ENIT and AHCD, along with synthetically generated data to improve generalization and class balance. The system includes a robust preprocessing pipeline consisting of denoising, binarization, normalization, and skew correction, enabling it to process both scanned and camera-captured images and generate editable outputs in text and PDF formats. Performance is evaluated using standard OCR metrics, including Character Error Rate (CER) and Word Error Rate (WER), with expected results of approximately 4.1% CER and 8.5% WER, outperforming traditional OCR systems like Tesseract. This work highlights the effectiveness of Transformer-based approaches for complex handwriting recognition tasks and contributes to the advancement of Arabic document digitization, with potential applications in domains such as government, education, healthcare, and digital archiving. This work was conducted at Arab International University (AIU), Syria. The official website of the university is: https://www.aiu.edu.sy