Chinese Journal of Liquid Crystals and Displays, Volume. 38, Issue 11, 1554(2023)

Underwater image enhancement method combining feature fusion and physical correction

De-xing WANG, Yu-rui YANG, Hong-chun YUAN*, and Kai GAO
Author Affiliations
  • School of Information,Shanghai Ocean University,Shanghai 201306,China
  • show less
    Figures & Tables(15)
    Overall network architecture
    Improved feature fusion module
    Residual module
    Improved attention module
    Flowchart of the physical enhancement module
    RGB pixel distribution line diagram
    Pixel distribution line diagram of the green channel
    Pixel distribution of each color channel.(a)Green channel pixel distribution map;(b)Blue channel pixel distribution map;(c)Red channel pixel distribution map.
    Qualitative comparison of ablation experiments.(a)Underwater image;(b)First experiment;(c)Second experiment;(d)Third experiment;(e)Reference image.
    Qualitative comparison of different methods in group A test set.(a)Underwater image;(b)UDCP;(c)Water-Net;(d)DBFusion;(e)Global-Local Net;(f)Ucolor;(g)LAFFNet;(h)Our method;(i)Reference images.
    Qualitative comparison of different methods for group B test set.(a)Underwater image;(b)UDCP;(c)Water-Net;(d)DBFusion;(e)Global-Local Net;(f)Ucolor;(g)LAFFNet;(h)Our method.
    • Table 1. Ablation experiment data table

      View table
      View in Article

      Table 1. Ablation experiment data table

      Test groupFirst testSecond testThird test
      NRMSE0.221 50.263 60.203 6
      PSNR21.305 819.808 322.140 7
      SSIM0.768 30.771 70.797 8
      UCIQE0.376 80.450 10.479 1
      IE6.964 37.209 67.381 9
      NIQE13.300 413.585 513.121 3
    • Table 2. Index comparison of enhanced images in group A test set

      View table
      View in Article

      Table 2. Index comparison of enhanced images in group A test set

      MethodsNRMSEPSNRSSIM
      UDCP0.413 815.727 10.644 4
      Water-Net0.275 619.429 70.760 9
      DBFusion0.388 116.198 70.676 2
      Global-Local Net0.259 620.637 90.780 1
      Ucolor0.280 719.764 40.740 3
      LAFFNet0.222 921.347 810.767 0
      Our method0.203 622.140 70.797 9
    • Table 3. Index comparison of enhanced images in group B test set

      View table
      View in Article

      Table 3. Index comparison of enhanced images in group B test set

      MethodsUCIQEIENIQE
      UDCP0.549 56.714 235.964 0
      Water-Net0.453 37.202 233.156 7
      DBFusion0.520 96.930 332.924 9
      Global-Local Net0.490 87.265 039.121 1
      Ucolor0.430 47.288 534.803 4
      LAFFNet0.412 27.076 636.773 1
      Our method0.584 37.485 331.494 6
    • Table 4. Comparison of the complexity of each method

      View table
      View in Article

      Table 4. Comparison of the complexity of each method

      MethodsParameters/MFLOPs/G
      Water-Net1.0422.87
      Global-Local Net0.008 00.000 02
      Ucolor102.62775.47
      LAFFNet0.159.77
      Our method0.035.95
    Tools

    Get Citation

    Copy Citation Text

    De-xing WANG, Yu-rui YANG, Hong-chun YUAN, Kai GAO. Underwater image enhancement method combining feature fusion and physical correction[J]. Chinese Journal of Liquid Crystals and Displays, 2023, 38(11): 1554

    Download Citation

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

    Category: Research Articles

    Received: Dec. 19, 2022

    Accepted: --

    Published Online: Nov. 29, 2023

    The Author Email: Hong-chun YUAN (hcyuan@shou.edu.cn)

    DOI:10.37188/CJLCD.2022-0382

    Topics