Scholarlink

<p>This paper addresses the challenges of single-image super-resolution (SISR) by introducing RPCAN (Residual Parallel Channel Attention Network). We suggest an innovative channel attention mechanism in the proposed model to enhance the quality of constructed super-resolution image. Traditional approaches primarily rely on Global Average Pooling (GAP) to weight the importance of channels based on their average values; however, this strategy does not properly pay attention to essential features such as edges and textures that are critical for effective reconstruction. Our proposed method integrates channel Contrast alongside GAP, allowing the model to prioritize channels rich in texture and detail, which significantly improves the recovery of fine details in super-resolution tasks. RPCAN utilizes a dual-branch architecture in the channel attention module, providing balanced emphasis on both GAP and Contrast measures. We rigorously evaluated RPCAN on five benchmark datasets across multiple scaling factors and different degradation models. We used objective metrics such as Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index (SSIM), along with qualitative visual assessments. Results demonstrate that RPCAN significantly outperforms state-of-the-art methods, achieving superior image reconstruction quality. These findings highlight the efficacy of combining channel contrast with GAP as a robust measure for channel attention. By integrating these two functions in a balanced architecture, RPCAN enhances performance while maintaining efficiency across various scenarios. The code implementation is publicly available at: <a href="https://github.com/AminTolou/RPCAN">https://github.com/AminTolou/RPCAN</a>.</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

Single image super resolution using parallel channel attention based on RCAN method (RPCAN)

  • 期刊论文

  • 发表日期：

    2025年7月11日

查看全文
  • Mohammad Amin Tolou Beydokhti,
  • Alireza Ahmadyfard,
  • Hossein Khosravi

摘要

This paper addresses the challenges of single-image super-resolution (SISR) by introducing RPCAN (Residual Parallel Channel Attention Network). We suggest an innovative channel attention mechanism in the proposed model to enhance the quality of constructed super-resolution image. Traditional approaches primarily rely on Global Average Pooling (GAP) to weight the importance of channels based on their average values; however, this strategy does not properly pay attention to essential features such as edges and textures that are critical for effective reconstruction. Our proposed method integrates channel Contrast alongside GAP, allowing the model to prioritize channels rich in texture and detail, which significantly improves the recovery of fine details in super-resolution tasks. RPCAN utilizes a dual-branch architecture in the channel attention module, providing balanced emphasis on both GAP and Contrast measures. We rigorously evaluated RPCAN on five benchmark datasets across multiple scaling factors and different degradation models. We used objective metrics such as Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index (SSIM), along with qualitative visual assessments. Results demonstrate that RPCAN significantly outperforms state-of-the-art methods, achieving superior image reconstruction quality. These findings highlight the efficacy of combining channel contrast with GAP as a robust measure for channel attention. By integrating these two functions in a balanced architecture, RPCAN enhances performance while maintaining efficiency across various scenarios. The code implementation is publicly available at: https://github.com/AminTolou/RPCAN.