Optics and Precision Engineering, Volume. 31, Issue 18, 2687(2023)

Global and local feature fusion image dehazing

Xin JIANG*... Haitao NIE and Ming ZHU |Show fewer author(s)
Author Affiliations
  • Changchun Institute of Optics, Fine Mechanics and Physics,Chinese Academy of Sciences, Changchun130033, China
  • show less
    Figures & Tables(13)
    Diagram of conditional generative adversarial network
    Framework of the generator
    Diagram of global and local feature fusion module
    Framework of enhancer
    Framework of the positional encoding generator
    Visual contrast effect on synthetic image datasets
    Visual contrast effect on real image datasets
    Visual contrast effect on real outdoor hazy images
    • Table 1. Parameter information of the generator

      View table
      View in Article

      Table 1. Parameter information of the generator

      网络层

      输出

      分辨率

      输出维度操作
      输入图像256×2563——
      Conv+BN+GELU256×256643×3 conv
      MaxPool128×128642×2 maxpool
      Conv+BN+GELU128×1281283×3 conv
      MaxPool64×641282×2 maxpool
      Conv+BN+GELU64×642563×3 conv
      MaxPool32×322562×2 maxpool
      Conv+BN+GELU32×325123×3 conv
      GLFFM32×32512详见2.3节
      Pixel Shuffle64×642562倍上采样
      Pixel Shuffle128×1281282倍上采样
      Pixel Shuffle256×256642倍上采样
      Enhancer256×256128详见2.4节
      Conv+Tanh256×25633×3 conv
    • Table 2. Objective evaluation results on synthetic image datasets

      View table
      View in Article

      Table 2. Objective evaluation results on synthetic image datasets

      去雾算法PSNRSSIM
      DCP17.692 10.877 8
      CAP18.475 80.820 4
      AODNet19.784 20.869 8
      EPDN21.503 60.879 2
      Pix2pix27.153 40.929 3
      FFA-Net31.126 50.958 3
      LD-Net24.765 30.917 2
      Ours33.190 20.977 0
    • Table 3. Objective evaluation results on real image datasets

      View table
      View in Article

      Table 3. Objective evaluation results on real image datasets

      去雾算法PSNRSSIM
      DCP17.982 30.698 2
      CAP17.104 60.656 4
      AODNet17.119 30.613 2
      EPDN17.216 80.708 1
      Pix2pix18.109 20.710 2
      FFA-Net17.658 20.691 0
      LD-Net17.132 80.632 5
      Ours19.311 50.747 8
    • Table 4. Objective evaluation results of different positional encoding methods

      View table
      View in Article

      Table 4. Objective evaluation results of different positional encoding methods

      编码方式合成图像数据集真实图像数据集
      PSNRSSIMPSNRSSIM
      全局位置编码生成器33.190 20.977 019.315 50.747 8
      局部位置编码生成器32.677 60.971 519.073 90.727 9
      固定位置编码32.698 20.972 518.977 30.724 7
      相对位置编码31.814 80.966 818.705 80.711 7
      可学习位置编码32.679 20.970 718.714 60.720 3
      无位置编码31.498 50.958 618.668 50.708 5
    • Table 5. Objective evaluation results of the enhancer

      View table
      View in Article

      Table 5. Objective evaluation results of the enhancer

      是否包含

      增强模块

      合成图像数据集真实图像数据集
      PSNRSSIMPSNRSSIM
      33.190 20.977 019.311 50.747 8
      32.486 90.970 117.534 20.662 7
    Tools

    Get Citation

    Copy Citation Text

    Xin JIANG, Haitao NIE, Ming ZHU. Global and local feature fusion image dehazing[J]. Optics and Precision Engineering, 2023, 31(18): 2687

    Download Citation

    EndNote(RIS)BibTexPlain Text
    Save article for my favorites
    Paper Information

    Category: Information Sciences

    Received: Feb. 13, 2022

    Accepted: --

    Published Online: Oct. 12, 2023

    The Author Email: JIANG Xin (xinjiang@zju.edu.cn)

    DOI:10.37188/OPE.20233118.2687

    Topics