Leveraging Arabic Text Embedded in Images: Challenges and Opportunities in NLP Analysis


  • AWS ABU EID Arab Apen University-JORDAN https://orcid.org/0000-0002-3974-8804
  • Achraf Ben Miled
  • Ashraf F. A. Mahmoud
  • Faroug A. Abdalla
  • Chams Jabnoun
  • Aida Dhibi
  • Firas M. Allan
  • Mohammed Ahmed Elhossiny
  • Imen Ben Mohamed
  • Marwa A. I. Elghazawy
  • Majid A. Nawaz
  • Salem Belhaj


Image Caption in Arabic, deep learning, text recognition, NLP


While recent advances in scene text recognition have blossomed, research has primarily focused on languages utilizing Latin scripts, neglecting languages with unique characteristics like Arabic. This study aims to bridge this gap by delving into the under-researched domain of Arabic scene text recognition. Describing Arabic images necessitates a fusion of computer vision and natural language processing, highlighting the intricate challenges AI algorithms encounter within this cross-domain, multi-modal landscape. The objective is to generate natural language descriptions for given test images, capturing crucial details such as characters, settings, actions, and more, while adhering to natural language conventions. However, the lack of readily available open-source Arabic datasets presents a significant obstacle, as most image description research revolves around English resources. Additionally, the inherent syntactic flexibility and linguistic nuances of Arabic descriptions amplify the algorithmic implementation challenges. Consequently, research concerning image descriptions, particularly in Arabic, needs to be explored more. To bridge this gap and facilitate further research, we introduce a novel dataset, the Arabic-English Daily Life Scene Text Dataset (EvArEST). Our study demonstrates promising progress in Arabic scene text recognition, highlighting both the challenges and opportunities of multi-modal AI algorithms. We conclude by emphasizing the need for more extensive datasets and algorithmic refinements to unlock the full potential of Arabic image descriptions in the context of NLP analysis.