Opto-Electronic Engineering, Volume. 51, Issue 9, 240142-1(2024)

Dual low-light images combining color correction and structural information enhance

Shanling Lin1...2, Yan Chen1,2, Xue Zhang1,2, Zhixian Lin1,2,3, Jianpu Lin1,2,*, Shanhong Lv1,2, and Tailiang Guo1,23 |Show fewer author(s)
Author Affiliations
  • 1School of Advanced Manufacturing, Fuzhou University, Quanzhou, Fujian 362200, China
  • 2Fujian Science & Technology Innovation Laboratory for Optoelectronic Information of China, Fuzhou, Fujian 350116, China
  • 3School of Physics and Information Engineering, Fuzhou University, Fuzhou, Fujian 350116, China
  • show less
    Figures & Tables(13)
    Schematic of generating the adversarial network
    Overall framework of U-Net
    Overall framework of the generator
    Detailed process of IGCB
    Overall framework of CAB
    Overall framework of SKCF
    Overall framework of the discriminator
    Ablation experiment visualization
    Visual comparison of different methods on LOL and LSRW
    Visual comparison of different methods on unreferenced datasets
    • Table 1. Different module ablation tests

      View table
      View in Article

      Table 1. Different module ablation tests

      模型CABSKCF上分支SDBPSNRSSIM
      F0××××15.660.498
      F1×××17.810.621
      F2×××17.470.678
      F3×××18.750.693
      F4×××17.130.623
      F5×18.470.615
      F6×19.120.644
      F7×18.250.613
      F8×19.310.669
      Ours19.890.672
    • Table 2. Objective evaluation results with mainstream methods

      View table
      View in Article

      Table 2. Objective evaluation results with mainstream methods

      MethodTypeLOLv1_testLOLv2_testLSRW-huaweiLSRW-Nikon
      PSNRSSIMLPIPSPSNRSSIMLPIPSPSNRSSIMLPIPSPSNRSSIMLPIPS
      Retinex-Net[7]S17.560.6040.47417.680.5940.44016.760.5660.43213.840.3920.352
      URetinex-Net[28]S17.480.6250.18218.160.6160.15612.520.4540.36912.510.4540.256
      EnGAN[9]U15.660.4980.35417.340.5940.31018.040.6070.36215.320.4900.267
      SCI[10]U13.810.5170.33917.250.5630.32215.100.4730.38115.280.4460.244
      Zero-DCE[8]U12.180.4770.33514.870.5090.31816.350.5640.36615.030.4480.239
      RUAS[29]U12.770.3340.41813.120.3290.37412.540.3620.44313.800.4060.287
      PairLIE[11]U18.470.6470.24519.880.7770.23418.980.5630.38715.520.4370.257
      PIE[12]U13.540.4850.35314.040.5370.32517.460.5880.41515.930.5440.298
      OursU19.890.6720.23420.080.6930.24218.570.6230.38317.030.5590.239
    • Table 3. NIQE values of different methods on each dataset

      View table
      View in Article

      Table 3. NIQE values of different methods on each dataset

      MethodMEFLIMENPEDICMALL
      Retinex-Net5.3396.8605.5425.3855.782
      URetinex-Net4.9765.1976.9745.2395.597
      EnGAN4.3104.5715.0204.5474.612
      Zero-DCE4.1024.8844.6354.3754.474
      SCI4.5815.4874.9785.0095.014
      RUAS4.5224.7806.3185.1855.201
      PairLIE4.5304.7354.5634.5294.589
      PIE4.1664.3224.5094.3614.340
      Ours4.0634.1424.5994.5024.327
    Tools

    Get Citation

    Copy Citation Text

    Shanling Lin, Yan Chen, Xue Zhang, Zhixian Lin, Jianpu Lin, Shanhong Lv, Tailiang Guo. Dual low-light images combining color correction and structural information enhance[J]. Opto-Electronic Engineering, 2024, 51(9): 240142-1

    Download Citation

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

    Category: Article

    Received: Jun. 17, 2024

    Accepted: Aug. 18, 2024

    Published Online: Dec. 12, 2024

    The Author Email:

    DOI:10.12086/oee.2024.240142

    Topics