Etudiant : ERRAJI FATIMA ZAHRAE

Filière : Master Informatique Décisionnelle et Vision Intelligente (MIDVI)

Encadrant : Pr. LOQMAN CHAKIR

Annèe : 2023

Résumé : Recent advancements in deep learning, computer vision, and natural language processing have revolutionized the way machines interpret unstructured data, bringing about significant changes. Optical Character Recognition (OCR) is a field that combines computer vision and natural language processing. Cursive writing in Arabic belongs to a non-Latin family that also includes Urdu, Chinese, and Hindi. Detecting and localizing Arabic writing in natural scene photos poses challenges, especially in identifying specific ligatures within the scene. The objective of this project is to develop Arabic Optical Character Recognition (AOCR). AOCR primarily involves two functions: segmentation and classification. However, segmentation in AOCR presents a substantial research obstacle. Arabic text is divided into three levels of segmentation: line, pseudoword, and character. Projections and other techniques are utilized to sequentially segment these three levels. Subsequently, a convolutional neural network is employed to extract features and recognize the segmented characters. The implementation of this project has yielded promising results, although it has acknowledged certain issues with the Arabic language and the absence of a standardized character database. This research proposes a methodology for detecting and identifying Arabic ligatures in visual media. The work includes the preparation of datasets, training the YOLO v8 with optimized parameters, performing regression analysis on the coordinate parameters and bounding box categories, and obtaining detection and recognition results. Keywords: OCR, Arabic, Segmentation, AOCR ,Classification, Recognition, Character, YOLOv8 , Ligature, detection