• Optics and Precision Engineering
  • Vol. 30, Issue 22, 2939 (2022)
Guoming YUAN1, Guang YANG2,*, Jinfeng WANG2, Haijun LIU1, and Wei WANG2
Author Affiliations
  • 1Department of Emergency Management, Institute of Disaster Prevention, Sanhe06520, China
  • 2Department of Information Engineering, Institute of Disaster Prevention, Sanhe06501, China.
  • show less
    DOI: 10.37188/OPE.20223022.2939 Cite this Article
    Guoming YUAN, Guang YANG, Jinfeng WANG, Haijun LIU, Wei WANG. Coarse-to-fine underwater image enhancement based on multi-level wavelet transform[J]. Optics and Precision Engineering, 2022, 30(22): 2939 Copy Citation Text show less
    Architecture of coarse-to-fine network for underwater image enhancement based on multi-level wavelet transform
    Fig. 1. Architecture of coarse-to-fine network for underwater image enhancement based on multi-level wavelet transform
    Comparison between initial images and enhanced images
    Fig. 2. Comparison between initial images and enhanced images
    High frequency image obtained by wavelet transform at different levels
    Fig. 3. High frequency image obtained by wavelet transform at different levels
    Architecture of RK2 block
    Fig. 4. Architecture of RK2 block
    Loss curves during training.
    Fig. 5. Loss curves during training.
    Average PSNR of testing data during training.
    Fig. 6. Average PSNR of testing data during training.
    Testing samples
    Fig. 7. Testing samples
    Comparison of enhanced results on synthetic underwater images
    Fig. 8. Comparison of enhanced results on synthetic underwater images
    Comparison of enhanced results on real underwater images
    Fig. 9. Comparison of enhanced results on real underwater images
    PSNR of different algorithms on Test1 and Test2.
    Fig. 10. PSNR of different algorithms on Test1 and Test2.
    LevelResIN_3ResIN_4ResIN_5
    PSNR24.3824.4724.50
    SSIM0.884 10.885 60.886 7
    ModelRes_3Res_4Res_6
    PSNR24.2324.4424.51
    SSIM0.882 50.884 90.886 8
    ModelRK2_3RK2_4RK2_6
    PSNR24.0124.1824.53
    SSIM0.874 30.879 20.886 8
    Table 1. Comparison of quantitative results by variant models with different number of building modules on Test 1.
    Leveln=1n=2n=3n=4n=5
    PSNR21.0823.3324.5124.4824.43
    SSIM0.830 40.866 70.886 70.881 70.876 4
    Modeln=6Net-w/o-lbNet-w/o-hbNet-w/o-RNet-Residual
    PSNR23.8922.4722.1521.7623.87
    SSIM0.871 10.841 20.839 70.838 90.869 5
    Table 2. Comparison of quantitative results by different variant models on Test1.
    ModelsEWTERHHUIEUWCNNUWGANWDNMWNFGANOurs
    PSNR21.0415.2511.4815.8712.1918.9823.6714.8424.51
    SSIM0.8360.7150.6350.7050.6480.7760.8640.7320.886
    PCQI0.9910.9830.9680.9740.9710.9871.0120.8921.057
    EI78.4574.1866.5271.5768.2976.4480.5764.0882.54
    UIQM2.9512.6162.3952.5532.4902.8213.0522.7953.123
    UCIQE0.6100.5940.5660.5810.5730.6050.6140.6080.625
    Entropy7.5287.2467.0297.1267.0977.4827.7147.3037.854
    Table 3. Quantitative comparison of enhanced results by different algorithms on Test1.
    ModelsEWTERHHUIEUWCNNUWGANWDNMWNFGANOurs
    PSNR17.6417.8418.5114.2317.3217.9618.2217.8120.18
    SSIM0.8230.8470.8030.6570.8220.7800.8090.7300.862
    PCQI0.9891.0180.9860.9690.9770.9800.9810.9831.035
    EI76.2077.6574.8968.4770.8973.4975.1275.8980.44
    UIQM2.8712.9212.8272.4562.7102.7332.8022.8142.987
    UCIQE0.6070.6130.6050.5870.5980.6010.6090.6030.618
    Entropy7.3897.4577.3387.0207.1177.2487.3087.3257.528
    Table 4. Quantitative comparison of enhanced results by different algorithms on Test2.
    Guoming YUAN, Guang YANG, Jinfeng WANG, Haijun LIU, Wei WANG. Coarse-to-fine underwater image enhancement based on multi-level wavelet transform[J]. Optics and Precision Engineering, 2022, 30(22): 2939
    Download Citation