Scholarlink

<p>Shadow removal in document images is essential for improving visual quality and readability. To address the limitations of existing methods under complex shadow conditions, we propose SdocDiff, a diffusion model-based approach that integrates DSE residual blocks and an LFCN-based mask refinement module. By incorporating mask guidance into the diffusion process, SdocDiff enables joint optimization of shadow-free images and refined masks, enhancing both shadow removal performance and text preservation. Experimental results on public benchmarks demonstrate that SdocDiff outperforms state-of-the-art methods in terms of PSNR and SSIM. Moreover, generalization tests reveal its superior robustness and accuracy across varying shadow intensities, highlighting its effectiveness and potential for real-world document enhancement tasks.</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

Document Enhancement with Diffusion Models: A Novel Shadow Removal Approach

  • 期刊论文

  • 发表日期：

    2025年10月21日

查看全文
  • Leyan Wang,
  • Shaofei Wu,
  • Jun Liu,
  • Nan Zhou,
  • Yuxin Wang

摘要

Shadow removal in document images is essential for improving visual quality and readability. To address the limitations of existing methods under complex shadow conditions, we propose SdocDiff, a diffusion model-based approach that integrates DSE residual blocks and an LFCN-based mask refinement module. By incorporating mask guidance into the diffusion process, SdocDiff enables joint optimization of shadow-free images and refined masks, enhancing both shadow removal performance and text preservation. Experimental results on public benchmarks demonstrate that SdocDiff outperforms state-of-the-art methods in terms of PSNR and SSIM. Moreover, generalization tests reveal its superior robustness and accuracy across varying shadow intensities, highlighting its effectiveness and potential for real-world document enhancement tasks.