Journals Articles Conferences News About CLP
Submit a paper Email Alert
Search
Search
Articles
Journals
News
Advanced Search
Top Searches
2D MaterialsTransformation opticsQuantum PhotonicsSmall lasersSemiconductor UV PhotonicsSupersymmetric microring laser arrays

Laser & Optoelectronics Progress, Volume. 61, Issue 14, 1437003(2024)

Nonuniform Defogging Based on Hierarchical Weight Interaction and Laplacian Prior

Yonghua Tang1, Yanjun Meng1、*, Sen Lin2, Feifan Shi1, Zhipeng Zhang1, and Xingtong Liu1
Author Affiliations
  • 1School of Information Science and Engineering, Shenyang University of Technology, Shenyang 110870, Liaoning , China
  • 2School of Automation and Electrical Engineering, Shenyang Ligong University, Shenyang 110159, Liaoning , China
    • show less
    AbstractGet PDF(in Chinese)
    Figures&Tables (19)
    Equations (16)
    References (35)
    Cited By (1)
    Tools
    Paper Information
    Topics
    Figures & Tables(19)
    Diagram of ND-HWILP network framework

    Fig. 1. Diagram of ND-HWILP network framework

    Download full size
    Unit structure

    Fig. 2. Unit structure

    Download full size
    Hierarchical weight interaction module

    Fig. 3. Hierarchical weight interaction module

    Download full size
    Total receptive field aggregation module

    Fig. 4. Total receptive field aggregation module

    Download full size
    DWT down-sampling module

    Fig. 5. DWT down-sampling module

    Download full size
    DWT up-sampling module

    Fig. 6. DWT up-sampling module

    Download full size
    Test results of different methods on O-Haze dataset

    Fig. 7. Test results of different methods on O-Haze dataset

    Download full size
    Test results of different methods on NH Haze dataset

    Fig. 8. Test results of different methods on NH Haze dataset

    Download full size
    Test results of different methods on NH Haze2 dataset

    Fig. 9. Test results of different methods on NH Haze2 dataset

    Download full size
    Test results of different methods on Dense Haze dataset

    Fig. 10. Test results of different methods on Dense Haze dataset

    Download full size
    Ablation results on NH Haze dataset

    Fig. 11. Ablation results on NH Haze dataset

    Download full size
    Object detection experiment

    Fig. 12. Object detection experiment

    Download full size
    Canny operator experiment

    Fig. 13. Canny operator experiment

    Download full size
    Feature matching experiment

    Fig. 14. Feature matching experiment

    Download full size

    • Table 1. Model parameter settings

      View table

      Table 1. Model parameter settings

      DatasetInitial learning rateLearning rate updateAdam parameterBatch sizeNumber of Layer modulesNumber of Unit modules
      O-Haze0.0001β1=0.9， β2=0.999， eps=1×10-8838
      NH Haze0.0010Half the learning rate for every 5 epochsβ1=0.9， β2=0.999， eps=1×10-8838
      NH Haze20.0010Half the learning rate for every 5 epochsβ1=0.9， β2=0.999， eps=1×10-8838
      Dense Haze0.0010Half the learning rate for every 2 epochsβ1=0.9， β2=0.999， eps=1×10-8838

    • Table 2. PSNR and SSIM using different algorithms on non-uniform datasets

      View table

      Table 2. PSNR and SSIM using different algorithms on non-uniform datasets

      Defogging algorithmO-HazeNH HazeNH Haze2Dense Haze
      PSNR /dBSSIMPSNR /dBSSIMPSNR /dBSSIMPSNR /dBSSIM
      AOD-Net15.690.58112.290.49914.250.44213.140.414
      PFFNet18.760.67415.980.67516.500.64313.580.533
      GridDehaze21.430.88117.210.72617.810.78114.190.363
      GCA20.670.75917.790.67817.930.75815.900.524
      UHD20.700.75418.330.74918.030.78314.750.540
      DehazeFormer23.490.85518.110.73419.130.81614.340.444
      ND-HWILP24.290.92419.870.82420.270.86616.130.591

    • Table 3. LPIPS and NIQE using different algorithms on non-uniform datasets

      View table

      Table 3. LPIPS and NIQE using different algorithms on non-uniform datasets

      Defogging algorithmO-HazeNH HazeNH Haze2Dense Haze
      LPIPSNIQELPIPSNIQELPIPSNIQELPIPSNIQE
      AOD-Net0.2523.4420.3933.1920.3183.2280.6043.441
      PFFNet0.2913.8340.3103.8710.3603.9760.3494.076
      GridDehaze0.1493.7420.2493.9840.2373.6440.3013.579
      GCA0.0784.1070.2073.8920.2593.0610.3534.692
      UHD0.0963.9040.1553.7840.2013.6060.3723.628
      DehazeFormer0.1223.1520.1843.1830.1183.0580.5083.354
      ND-HWILP0.0543.0010.1242.9930.0593.0170.2853.295

    • Table 4. Running time and parameter quantity of different algorithms

      View table

      Table 4. Running time and parameter quantity of different algorithms

      MethodTime /sParams /MB
      AOD-Net0.0040.00167
      PFFNet0.32619.88700
      GridDehaze0.1880.91400
      GCA0.0110.67000
      UHD0.14332.94900
      DehazeFormer0.0191.22500
      ND-HWILP0.0382.28300

    • Table 5. Ablation experiment

      View table

      Table 5. Ablation experiment

      ModelPSNR /dBSSIM
      Base17.30170.6998
      Base+C18.12250.7621
      Base+C+E18.26130.7764
      Base+C+E+L19.00390.8177
      Base+C+E+L+W19.86870.8243
    Tools

    Get Citation

    Copy Citation Text

    Yonghua Tang, Yanjun Meng, Sen Lin, Feifan Shi, Zhipeng Zhang, Xingtong Liu. Nonuniform Defogging Based on Hierarchical Weight Interaction and Laplacian Prior[J]. Laser & Optoelectronics Progress, 2024, 61(14): 1437003

    Download Citation

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

    Category: Digital Image Processing

    Received: Oct. 9, 2023

    Accepted: Oct. 30, 2023

    Published Online: Jul. 8, 2024

    The Author Email: Yanjun Meng (mengyanjun2021@163.com)

    DOI:10.3788/LOP232261

    CSTR:32186.14.LOP232261

    Topics

    laser devices and laser physics
    Lasers and Laser Optics
    Laser physics
    laser manufacturing
    Instrumentation, Measurement and Metrology