Scholarlink

<p>Images captured on rainy days often pose challenges for computer vision tasks, as rain streaks interfere and produce blurred images. Existing rain removal algorithms often fail to recover the lost image details effectively. To address these problems, we propose a novel two-stage progressive residual learning network, which consists of two networks: a rain streaks removal network and a detail progressive repair network. The rain streaks removal network employs rain residual modules, which combine an attention mechanism with residual learning to better capture spatial contextual information and efficiently remove rain streaks. In the detail progressive repair network, we design shallow and deep detail repair modules to decompose this complex repair task into simpler steps, thereby enhancing repair capability and reducing information loss. Experimental results demonstrate that our method achieves superior performance on synthetic rain image datasets, including Rain200L, Rain200H, and Rain800. It attains Structural Similarity Index Measure (SSIM) values of 0.988, 0.967, and 0.926, and Peak Signal-to-Noise Ratio (PSNR) values of 39.365 dB, 30.352 dB, and 29.351 dB, respectively. Our method also performs well on real-world rainy images, not only removing rain streaks of varying sizes but also preserving background details, thereby outperforming other state-of-the-art methods in both visual perception and quantitative evaluation.</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

Two-stage progressive residual learning network for rain streak removal

  • 期刊论文

  • 发表日期：

    2025年6月30日

查看全文
  • Mengzhao Liu,
  • Sen Lin,
  • Yonghui Hou,
  • Shuming Zhang

摘要

Images captured on rainy days often pose challenges for computer vision tasks, as rain streaks interfere and produce blurred images. Existing rain removal algorithms often fail to recover the lost image details effectively. To address these problems, we propose a novel two-stage progressive residual learning network, which consists of two networks: a rain streaks removal network and a detail progressive repair network. The rain streaks removal network employs rain residual modules, which combine an attention mechanism with residual learning to better capture spatial contextual information and efficiently remove rain streaks. In the detail progressive repair network, we design shallow and deep detail repair modules to decompose this complex repair task into simpler steps, thereby enhancing repair capability and reducing information loss. Experimental results demonstrate that our method achieves superior performance on synthetic rain image datasets, including Rain200L, Rain200H, and Rain800. It attains Structural Similarity Index Measure (SSIM) values of 0.988, 0.967, and 0.926, and Peak Signal-to-Noise Ratio (PSNR) values of 39.365 dB, 30.352 dB, and 29.351 dB, respectively. Our method also performs well on real-world rainy images, not only removing rain streaks of varying sizes but also preserving background details, thereby outperforming other state-of-the-art methods in both visual perception and quantitative evaluation.