This Author published in this journals
All Journal Technomedia Journal
Claim Missing Document
Check
Articles

Found 1 Documents
Search

Attention Enhanced Convolutional Neural Networks for Arabic Calligraphy Recognition Indra Riyanto; Muhamad Rido; Marta Rodriguez
Technomedia Journal Vol 11 No 2 (2026): October
Publisher : Pandawan Incorporation, Alphabet Incubator Universitas Raharja

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.33050/g9wzdk67

Abstract

Arabic calligraphy recognition remains a challenging task in computer vision due to the cursive nature of Arabic characters, complex ligatures, diacritics, and diverse artistic writing styles. These characteristics create difficulties for automatic classification systems, particularly in distinguishing visually similar calligraphy patterns. Recent advances in deep learning, especially Convolutional Neural Networks (CNN) and attention mechanisms, provide opportunities to improve feature representation for complex visual recognition tasks. This study aims to develop an Arabic calligraphy recognition model by integrating CNN with the Convolutional Block Attention Module (CBAM) to improve the classification performance of three Arabic calligraphy styles: Naskhi, Riq’ah, and Diwani. The proposed approach utilizes a dataset consisting of 946 Arabic calligraphy images. The images undergo preprocessing, normalization, and augmentation before being processed using CNN for hierarchical feature extraction. CBAM is then applied to enhance feature representation through channel attention and spatial attention mechanisms. The proposed model is evaluated using accuracy, precision, recall, and F1-score metrics. Experimental results demonstrate that the CNN-CBAM model achieves a validation accuracy of 87.37%, outperforming the baseline CNN model with 58.42%. The proposed model also obtains recall values of 100.00% for Riq’ah, 95.31% for Diwani, and 68.18% for Naskhi. The integration of CNN and CBAM effectively improves Arabic calligraphy classification by enhancing discriminative feature learning. The pro- posed approach provides potential support for digital preservation, automated indexing, and intelligent management of Arabic calligraphy collections.