Scholarlink

<p>Despite significant advancements in handwriting analysis, writer identification in Arabic manuscripts remains challenging due to the scarcity of large-scale annotated datasets. Traditional methods, which primarily rely on supervised learning and convolutional architectures, often struggle in such low-resource conditions. In contrast, unlabeled Arabic handwritten data is often available in large quantities. To address this gap, we propose in this paper an end-to-end Vision Transformer (ViT)-based framework for Arabic writer identification. Our approach leverages self-supervised learning during pretraining to acquire robust and transferable feature representations from unlabeled handwritten words, followed by supervised fine-tuning phase for writer identification task. Additionally, we integrate a synthetic data generation strategy to further mitigate data scarcity issues. The proposed framework achieves state-of-the-art performance on two benchmark Arabic datasets: the IFN/ENIT and the AHTID/MW datasets.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

ST-WID: Self-supervised transformer for writer identification in arabic handwritten scripts

  • 期刊论文

  • 发表日期：

    2025年9月22日

查看全文
  • Islem Fatnassi,
  • Sana Khamekhem Jemni,
  • Sourour Ammar,
  • Yousri Kessentini

摘要

Despite significant advancements in handwriting analysis, writer identification in Arabic manuscripts remains challenging due to the scarcity of large-scale annotated datasets. Traditional methods, which primarily rely on supervised learning and convolutional architectures, often struggle in such low-resource conditions. In contrast, unlabeled Arabic handwritten data is often available in large quantities. To address this gap, we propose in this paper an end-to-end Vision Transformer (ViT)-based framework for Arabic writer identification. Our approach leverages self-supervised learning during pretraining to acquire robust and transferable feature representations from unlabeled handwritten words, followed by supervised fine-tuning phase for writer identification task. Additionally, we integrate a synthetic data generation strategy to further mitigate data scarcity issues. The proposed framework achieves state-of-the-art performance on two benchmark Arabic datasets: the IFN/ENIT and the AHTID/MW datasets.