Scholarlink

<p>Single-image dehazing plays a critical role in enhancing complex computer vision tasks, such as object tracking, recognition and detection, especially in adverse weather conditions. Haze reduction improves visual clarity and quality, leading to more accurate downstream processing. However, many existing methods fail to account for the intricate blending of haze and background, leading to feature redundancy and suboptimal restoration in practical applications. To address these challenges, we propose a Novel, lightweight four-stage architecture referred as Multiscale Feature Recomposition Network (MSFRNet) for restoring haze-free images. Our network integrates two key components: Twin Prior Fusion Module (TPFM) and Pyramid Dilated Split Attention Unit (PDSAU). The TPFM integrates multiple sources of prior information, enhancing structural integrity and preserving fine details, while the PDSAU efficiently disentangles haze from important image features and recombines them for accurate image dehazing. Our network demonstrates significant performance improvements in terms of both accuracy and speed when tested on synthetic and real-world datasets, and outperformsover the existing state-of-the-art networks. With a runtime of just 7&#xa0;ms and the designed lightweight architecture uses 2.24&#xa0;million parameters, which showcases our network is highly efficient for real-time, and resource-constrained applications.</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

MSFRNet: Multiscale Feature Recomposition Network for SingleImage Dehazing

  • 期刊论文

  • 发表日期：

    2025年5月7日

查看全文
  • D. Pushpalatha,
  • P. Prithvi

摘要

Single-image dehazing plays a critical role in enhancing complex computer vision tasks, such as object tracking, recognition and detection, especially in adverse weather conditions. Haze reduction improves visual clarity and quality, leading to more accurate downstream processing. However, many existing methods fail to account for the intricate blending of haze and background, leading to feature redundancy and suboptimal restoration in practical applications. To address these challenges, we propose a Novel, lightweight four-stage architecture referred as Multiscale Feature Recomposition Network (MSFRNet) for restoring haze-free images. Our network integrates two key components: Twin Prior Fusion Module (TPFM) and Pyramid Dilated Split Attention Unit (PDSAU). The TPFM integrates multiple sources of prior information, enhancing structural integrity and preserving fine details, while the PDSAU efficiently disentangles haze from important image features and recombines them for accurate image dehazing. Our network demonstrates significant performance improvements in terms of both accuracy and speed when tested on synthetic and real-world datasets, and outperformsover the existing state-of-the-art networks. With a runtime of just 7 ms and the designed lightweight architecture uses 2.24 million parameters, which showcases our network is highly efficient for real-time, and resource-constrained applications.