Laser & Optoelectronics Progress, Volume. 62, Issue 15, 1500009(2025)

Research Advances on Unsupervised Networks-Driven Imaging Through Scattering Media (Invited)

Longyu Qiao1, Bing Lin1, Xueqiang Fan1, Xixun Sun1,2, Zhiyong Peng2, and Zhongyi Guo1、*
Author Affiliations
  • 1School of Computer and Information, Hefei University of Technology, Hefei 230601, Anhui , China
  • 2Tianjin Jinhang Institute of Technical Physics, Tianjin 300192, China
  • show less
    Figures & Tables(16)
    Underwater scattering environment and atmospheric scattering environment
    Algorithms for AE-based IUL methods. (a1)(a2) Network structure of underwater image and video enhancement framework based on convolutional AE and its comparative results with different methods[51]; (b1)(b2) network structure of Retinex algorithm enhanced framework based on fractional integration and comparison results with different methods[52]; (c) network structure of UDNet enhancement framework[53]; (d) network structure of the RQUNet-VAE method[54]
    Different algorithms for GAN-based IUL methods. (a1)(a2) Network structure of unsupervised underwater visual enhancement methods and the results of underwater visual enhancement produced by their outputs[62]; (b1)(b2) network structure of low-light visual enhancement methods and comparative experimental results of different methods[63]; (c) network structure of MAGAN[64]; (d) the method of attention-guided dual-discriminator based GAN[65]
    Schematic diagram of CycleGAN [66]
    IUL methods based on different CycleGAN. (a1)(a2) CycleGAN-based target recovery method and model reconstruction results[67]; (b1)(b2) network structure combining cyclic consistency loss and perceptual loss and its comparison with CycleGAN[68]; (c) network structure of Cycle-SNSPGAN method[69]; (d) network structure of the UME-Net method[70]
    GAN based on physical models and other learning approaches. (a1)(a2) Network structure of the priori-free GAN incorporating autocorrelation consistency and model output results[71]; (b1)(b2) network structure of PDR-GAN and comparison results with other methods[72]; (c) network structure of EnlightenGAN[73]; (d1)(d2) flow of UCL-Dehaze and comparison results with other methods[74]; (e) PatchNCE network structure based on contrast learning [75]
    Concrete process of DDPM [78]
    DDPM-based IUL methods. (a1)(a2) Enhanced network model and its output results [79]; (b) new biconditional-based DDPM [80]; (c1)(c2) DDPM-based fusion method and comparison results of different road inspection methods [81]; (d1)(d2) hybrid model based on DDPM and atmospheric scattering model and results of the model dehazing and detection [82]
    CNN-based IUL methods. (a1)(a2) Network structure of dehazing algorithm based on interactive fusion module and iterative optimization module[84]; (b1)(b2) algorithm process of similarity fusion strategy and its comparison results with different methods[85]; (c1)(c2) network structure of unsupervised fusion method and its comparison results with different methods[86]; (d1)(d2) network structure of generating pseudo labels through bright channels and its comparison results with different methods[87]; (e1)(e2) specific algorithm process of synthesizing pseudo real images through multi-source images and its comparison results with different methods[89]; (f1)(f2) specific algorithm process for inferring the distribution of real data through estimating noise models and its comparison results with other methods[90]
    • Table 1. Performance analysis of IUL methods based on different AEs

      View table

      Table 1. Performance analysis of IUL methods based on different AEs

      Ref.PSNR↑SSIM↑LER↑CER↑UIQM↑UCIQE↑
      5118.7700.633
      5227.8020.8491.5831.955
      5322.9600.7713.2650.749
      5439.0870.971
    • Table 2. Performance analysis of IUL methods based on different GANs

      View table

      Table 2. Performance analysis of IUL methods based on different GANs

      Ref.PSNR↑SSIM↑BRISQUE↓NIQE↓EME↑EN↓SD↑MI↑
      621.2847
      6326.64510.881721.22184.4719
      6422.38950.847037.9784
      6571.16070.76694.147852.46403.0764
    • Table 3. Performance analysis of IUL methods based on different CycleGANs

      View table

      Table 3. Performance analysis of IUL methods based on different CycleGANs

      Ref.PSNR↑SSIM↑RMSE↓
      670.7700.17
      6819.920.640
      6929.200.964
      7021.560.745
    • Table 4. Performance analysis of IUL methods based on different physical models and other learning methods

      View table

      Table 4. Performance analysis of IUL methods based on different physical models and other learning methods

      Ref.PSNR↑SSIM↑MAE↓MSE↓NIQE↓BRISQUE↓SSEQ↓PI↓UCIQE↑UIQM↑
      7117.810.6640.0293
      7226.300.947
      7426.730.9473.73624.65826.0283.412
      7518.100.7101095.218.104.012
    • Table 5. Performance analysis of IUL method based on different diffusion models

      View table

      Table 5. Performance analysis of IUL method based on different diffusion models

      Ref.PSNR↑SSIM↑MSE↓LPIPS↓FID↓
      7931.09230.891414.5778
      8024.26000.176080.99
      8238.59000.09890.0040
    • Table 6. Performance analysis of IUL method based on different CNN algorithms

      View table

      Table 6. Performance analysis of IUL method based on different CNN algorithms

      Ref.PSNR↑SSIM↑UQI↑IE↓SD↑MI↑NIQE↑LOE↓
      8420.510.8000.865
      8523.580.926
      860.5187.10740.461.680
      873.30305
      8919.720.875
      9030.190.839
    • Table 7. Performance analysis of different typical IUL methods

      View table

      Table 7. Performance analysis of different typical IUL methods

      Ref.PSNR↑SSIM↑UIQM↑UCIQE↑EN↓SD↑MI↑MSE↓UQI↑
      5322.96000.77103.2650.749
      6571.16070.76694.147852.4643.0764
      7931.09230.891414.5778
      8420.51000.80000.865
    Tools

    Get Citation

    Copy Citation Text

    Longyu Qiao, Bing Lin, Xueqiang Fan, Xixun Sun, Zhiyong Peng, Zhongyi Guo. Research Advances on Unsupervised Networks-Driven Imaging Through Scattering Media (Invited)[J]. Laser & Optoelectronics Progress, 2025, 62(15): 1500009

    Download Citation

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

    Category: Reviews

    Received: Apr. 9, 2025

    Accepted: May. 13, 2025

    Published Online: Aug. 6, 2025

    The Author Email: Zhongyi Guo (guozhongyi@hfut.edu.cn)

    DOI:10.3788/LOP250975

    CSTR:32186.14.LOP250975

    Topics