Arabic Handwritten Letters Recognition Using Convolutional Neural Networks

Arabic handwriting recognition is a complex task due to the script’s cursive nature, the presence of diacritics, character positioning, connected letters, and variations in individual writing styles. This study focuses on the development of a handwritten Arabic letter recognition system using Convolutional Neural Networks (CNN). Therefore, recognition of handwritten letters is challenging task towards text recognition. However, CNN have emerged as a powerful tool to address this task, as they rely on the ability of CNN to learn features directly from the input image. This research involves two stages: creating an Arabic Handwritten Letters Dataset (AHLD) and constructing a CNN-based recognition model. The AHLD dataset created comprises 8,000 labeled images representing the 28 Arabic letters, preprocessed and structured. The CNN model is designed to automatically extract meaningful features and classify Arabic handwritten letters with high accuracy. Various CNN architectures were evaluated, with the best performing model achieving a classification accuracy of 97.17 %. The study highlights the challenges of Arabic script recognition and demonstrates the effectiveness of deep learning techniques in recognizing handwritten letters, while also contributing to improve handwritten Arabic text recognition. Additionally, the AHLD dataset is made publicly available to support future research in this field.