Scholarlink

<p>Most image dehazing algorithms in practical applications suffer from weak input adaptability due to their neglect of input image types (hazy or haze-free), and they often enhance dehazing performance by deepening network layers, resulting in large parameter counts that make real-time processing difficult. To address these issues, we propose a real-time image dehazing algorithm based on adaptive routing. Firstly, we design an adaptive routing module based on multi-scale feature extraction to determine the processing path of the input image. Then, based on the traditional feature attention (FA) module, a lightweight de-redundant feature attention (DRFA) module is constructed for dehazing processing. Finally, the residual network with the PixelShuffle upsampling module is utilized for efficient image recovery. The experimental results demonstrate that our algorithm achieves high PSNR and SSIM scores on several datasets, and we reach an average PSNR and SSIM of 26.75&#xa0;dB and 0.832 on the mixed dataset of clear and hazy images, respectively. These results outperform other comparative algorithms, indicating the robustness and practicality of our approach. Not only that, our algorithm has a computational cost of 8.93G FLOPs and a dehazing time of only 4.81&#xa0;ms during the 256 × 256-pixel image dehazing efficiency test, which meets the requirements of real-time processing for video surveillance in outdoor scenes.</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

A real-time image dehazing algorithm based on adaptive routing

  • 期刊论文

  • 发表日期：

    2025年6月12日

查看全文
  • Kun Meng,
  • Shengjie Yang,
  • Zhen Liu,
  • Sheng Li,
  • Min Wang,
  • Haibin Liao,
  • Li Yuan

摘要

Most image dehazing algorithms in practical applications suffer from weak input adaptability due to their neglect of input image types (hazy or haze-free), and they often enhance dehazing performance by deepening network layers, resulting in large parameter counts that make real-time processing difficult. To address these issues, we propose a real-time image dehazing algorithm based on adaptive routing. Firstly, we design an adaptive routing module based on multi-scale feature extraction to determine the processing path of the input image. Then, based on the traditional feature attention (FA) module, a lightweight de-redundant feature attention (DRFA) module is constructed for dehazing processing. Finally, the residual network with the PixelShuffle upsampling module is utilized for efficient image recovery. The experimental results demonstrate that our algorithm achieves high PSNR and SSIM scores on several datasets, and we reach an average PSNR and SSIM of 26.75 dB and 0.832 on the mixed dataset of clear and hazy images, respectively. These results outperform other comparative algorithms, indicating the robustness and practicality of our approach. Not only that, our algorithm has a computational cost of 8.93G FLOPs and a dehazing time of only 4.81 ms during the 256 × 256-pixel image dehazing efficiency test, which meets the requirements of real-time processing for video surveillance in outdoor scenes.