Opto-Electronic Engineering, Volume. 49, Issue 7, 210448(2022)

A generative adversarial network incorporating dark channel prior loss used for single image defogging

Deqiang Cheng1...2, Yangyang You2, Qiqi Kou3,* and Jinyang Xu2 |Show fewer author(s)
Author Affiliations
  • 1Engineering Research Center of Underground Space Intelligent Control, Ministry of Education, Xuzhou, Jiangsu 221000, China
  • 2School of Information and Control Engineering, China University of Mining and Technology, Xuzhou, Jiangsu 221000, China
  • 3School of Computer Science and Technology, China University of Mining and Technology, Xuzhou, Jiangsu 221000, China
  • show less
    Figures & Tables(15)
    Framework of adversarial generation network
    Dark channel feature comparison.(a) Original images; (b) Dark channel feature
    Dark channel feature intensity distribution. (a) Intensity distribution; (b) Average intensity distribution of 5000 images
    Framework of the proposed algorithm
    Qualitative comparison on synthetic images
    Qualitative comparison on real hazy images
    Quantitative comparison with control group on SOTS test-set & synthetic images of HSTS test-set
    Qualitative comparison with control groups on real images of HSTS test-set
    Quantitative comparison with control group on real hazy images of HSTS test-set
    • Table 1. Parameters of generator network

      View table
      View in Article

      Table 1. Parameters of generator network

      卷积层ConvResConvResConvResUpconvResUpconvResConvTanh
      输入通道数3646412812825625612812864643
      输出通道数6464128128256256128128646433
      卷积核尺寸753447
      步长122221
      边界填充311113
    • Table 2. Parameters of the discriminator network

      View table
      View in Article

      Table 2. Parameters of the discriminator network

      卷积层Conv1Conv2Conv3Conv4Conv5
      输出通道数641282565121
      卷积核大小44444
      步长22211
    • Table 3. Quantitative results of each algorithm on SOTS test-set & synthetic images of HSTS test-set

      View table
      View in Article

      Table 3. Quantitative results of each algorithm on SOTS test-set & synthetic images of HSTS test-set

      数据集HSTSSOTS-outdoorSOTS-indoor
      评价指标PSNRSSIMPSNRSSIMPSNRSSIM
      DCP[7]17.220.8017.560.8220.150.87
      BCCR[9]15.090.7415.490.7816.880.79
      CAP[10]21.540.8722.300.9119.050.84
      MSCNN[15]18.290.8419.560.8617.110.81
      D-Net[16]24.490.9222.720.8621.140.85
      AOD-Net[17]21.580.9221.340.9219.380.85
      GFN[18]22.940.8721.490.8422.320.88
      DEnergy[26]24.440.9324.080.9319.250.83
      本文算法25.350.9625.170.9623.700.82
    • Table 4. Quantitative results of each algorithm on D-HAZY & HazeRD & BeDDE test-set

      View table
      View in Article

      Table 4. Quantitative results of each algorithm on D-HAZY & HazeRD & BeDDE test-set

      数据集D-HAZYHazeRDBeDDE
      评价指标PSNRSSIMPSNRSSIMVSIVIRI
      DCP[7]15.090.8314.010.390.9460.9110.965
      MSCNN[15]13.570.8015.580.420.9470.8920.972
      D-Net[16]13.760.8115.530.410.9520.8900.972
      AOD-Net[17]13.130.7915.630.450.9540.8960.970
      CycleGAN[22]13.550.7715.640.440.9420.8660.961
      RefineDNet[39]15.440.8315.610.430.9600.9070.971
      SM-Net[24]15.320.8115.550.400.9610.8990.969
      本文算法15.390.8215.590.440.9670.8990.967
    • Table 5. Quantitative results of the control groups & proposed algorithm on real images of HSTS test-set

      View table
      View in Article

      Table 5. Quantitative results of the control groups & proposed algorithm on real images of HSTS test-set

      Pic1Pic2Pic3Pic4Pic5Pic6Pic7Pic8Pic9Pic10
      对照组1e0.0162-0.05060.08680.16290.38690.33910.62621.00140.95790.0160
      r1.01001.04351.12911.04501.67431.42861.46411.46061.82781.0928
      p0.00000.000000.001900.000200.00020.00030.0000
      对照组2e0.33600.64210.27090.07740.38330.43380.78451.83480.62070.4650
      r1.46051.42221.26401.20471.69261.43821.33791.51261.60211.1351
      p0.01020.11550.00010.008100.00720.00020.00810.00160.0286
      本文算法e0.35340.60610.88700.12320.45040.62651.23782.07141.14430.5197
      r1.62841.55761.66311.35722.15991.68631.57931.78141.95611.3026
      p0.00820.10150.00010.002000.00790.00110.01080.00130.0153
    • Table 6. Run time of each algorithm on SOTS test-set

      View table
      View in Article

      Table 6. Run time of each algorithm on SOTS test-set

      DCP[7]CAP[10]MSCNN[15]D-Net[16]AOD-Net[17]GFN[18]CycleGAN[22]本文算法
      DeviceCPUCPUGPUCPUGPUGPUGPUGPU
      Run time1.740.723.413.330.566.102.961.37
    Tools

    Get Citation

    Copy Citation Text

    Deqiang Cheng, Yangyang You, Qiqi Kou, Jinyang Xu. A generative adversarial network incorporating dark channel prior loss used for single image defogging[J]. Opto-Electronic Engineering, 2022, 49(7): 210448

    Download Citation

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

    Category: Article

    Received: Jan. 19, 2022

    Accepted: --

    Published Online: Aug. 1, 2022

    The Author Email: Kou Qiqi (1120074179@qq.com)

    DOI:10.12086/oee.2022.210448

    Topics