Scholarlink

<p>Ancient manuscript restoration and binarization are two very important steps in the process of analysis and recognition of ancient manuscripts to retain historical contents with better readability. We suggest a new model called RSEUNet (Residual Squeeze and Excitation U-Net), which can effectively restore and binarize historical documents by adding an advanced SE (Squeeze-and-Excitation) block to a U-Net architecture along with a ResNet encoder. These SE blocks change the channel-wise feature responses in a way that is more adaptive, which makes the network better at representing things. Our approach has been extensively tested on the DIBCO dataset. RSEUNet outperforms state-of-the-art methods in F-measure, pseudo-F-measure (p-FM), PSNR, and DRD, achieving a p-FM of 98.38% for DIBCO 2017 and 99.17% for DIBCO 2018. These results demonstrate the model’s effectiveness in enhancing historical document digitization and preservation.</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

RSEUNet: residual squeeze and excitation U-Net for restoration and binarization of historical documents

  • 期刊论文

  • 发表日期：

    2025年8月12日

查看全文
  • Sarra Hamza,
  • Rafik Menassel,
  • Chawki Djeddi

摘要

Ancient manuscript restoration and binarization are two very important steps in the process of analysis and recognition of ancient manuscripts to retain historical contents with better readability. We suggest a new model called RSEUNet (Residual Squeeze and Excitation U-Net), which can effectively restore and binarize historical documents by adding an advanced SE (Squeeze-and-Excitation) block to a U-Net architecture along with a ResNet encoder. These SE blocks change the channel-wise feature responses in a way that is more adaptive, which makes the network better at representing things. Our approach has been extensively tested on the DIBCO dataset. RSEUNet outperforms state-of-the-art methods in F-measure, pseudo-F-measure (p-FM), PSNR, and DRD, achieving a p-FM of 98.38% for DIBCO 2017 and 99.17% for DIBCO 2018. These results demonstrate the model’s effectiveness in enhancing historical document digitization and preservation.