Infrared and Laser Engineering, Volume. 51, Issue 9, 20210919(2022)

Depth-rectified statistical scattering modeling for deep-sea video descattering

Yihui Fan1, Xin Jin1, Rujia Deng1, Jiayu Xie1, Kelin Sun2, Jingchuan Yang2, and Bing Zhang2
Author Affiliations
  • 1Shenzhen Key Laboratory of Broadband Network and Multimedia,Shenzhen International Graduate School, Tsinghua University, Shenzhen 518055, China
  • 2Deep Sea Video Technology Laboratory, Institute of Deep-sea Science and Engineering, Chinese Academy of Sciences, Sanya 572000, China
  • show less
    Figures & Tables(9)
    Non-uniform illumination and multi-depth scenarios in deep-sea
    Overall framework of the proposed deep-sea descattering method
    Subjective comparisons on shallow sea and deep-sea images. From left to right are input underwater images, and the results of DCP[14], UDCP[16], Fusion[13], Retinex[9], GUM[10], Peplography[17], CLAHE[8], and the proposed method
    Objective comparisons of deep-sea video. (a)-(d) are the video frame samples in the video and their UCIQE values; (e) is the UCIQE results of the video
    Objective comparisons of six deep-sea videos corresponding to Deep 1-6. (a1) to (l1) are the video frame samples in the six videos; (a2)-(l2) are the results of (a1) to (l1) of the proposed algorithm; (m)-(r) are the UCIQE results of the six videos by different methods. The red dotted line in (m) to (r) mark the corresponding frame of the above video frame samples in (a1)-(l1)
    Subjective and objective comparisons on deep-sea images with different scattering intensities from Haze 1 to Haze 5. From left to right are input underwater images, and the results of DCP[14], UDCP[16], Fusion[13], Retinex[9], GUM[10], Peplography[17], CLAHE[8], and the proposed method
    • Table 1. Overview about 2 images and 6 video sequences

      View table
      View in Article

      Table 1. Overview about 2 images and 6 video sequences

      Video sequenceFrame rateResolutionFrame numberContentsScattering intensitySample images
      Shallow 11 fps640×4801Coral+
      Shallow 21 fps640×4801Fish+
      Deep 130 fps1 920×930300Robot arm and acquisition equipments+++
      Deep 230 fps1 920×9301 080Robot arm and acquisition equipments++
      Deep 330 fps1 920×930750Robot arm and acquisition equipments++++
      Deep 430 fps1 920×930450Seabed and plant+
      Deep 530 fps1 920×930420Seabed++++
      Deep 630 fps3 840×2 160330Seabed+
    • Table 2. UCIQE values of different methods in Fig.3 (The bold values represent the best results)

      View table
      View in Article

      Table 2. UCIQE values of different methods in Fig.3 (The bold values represent the best results)

      Input imagesDCP[14]UDCP[16]Fusion[13]Retinex[9]GUM[10]Peplography[17]CLAHE[8]Proposed
      Shallow 10.47110.57550.57450.51120.63970.52170.64130.53390.6803
      Shallow 20.36830.47890.47140.44370.62740.39520.56290.39570.6299
      Deep 10.52500.62130.58890.53580.57890.53980.48990.53110.6259
      Deep 20.44170.41370.38020.47650.53900.48360.37800.48140.5828
      Deep 30.43350.43160.41300.47710.55250.43870.34550.43710.6644
      Deep 40.34000.49170.43160.42040.59190.36400.52330.36170.6372
      Deep 50.34550.44640.39850.42270.55570.36610.33030.36680.6344
      Deep 60.59790.63920.62440.58870.56920.59660.32970.59080.6459
    • Table 3. Averaged UCIQE values of different methods with above 6 video sequences(The bold values represent the best results)

      View table
      View in Article

      Table 3. Averaged UCIQE values of different methods with above 6 video sequences(The bold values represent the best results)

      Video IDInput videoDCP[14]UDCP[16]Fusion[13]Retinex[9]GUM[10]Peplography[17]CLAHE[8]Proposed
      Deep 10.52100.61410.58320.53280.57550.53410.48780.48720.6267
      Deep 20.45380.42250.41400.48410.55620.49580.37670.39110.5829
      Deep 30.42540.42800.40640.47050.52660.42630.33620.46560.6586
      Deep 40.34970.47920.38450.42420.57420.38010.47360.40700.6261
      Deep 50.34390.46060.39600.42250.53310.35870.32540.41050.6256
      Deep 60.59800.63930.62440.58840.56920.59680.33190.55640.6450
    Tools

    Get Citation

    Copy Citation Text

    Yihui Fan, Xin Jin, Rujia Deng, Jiayu Xie, Kelin Sun, Jingchuan Yang, Bing Zhang. Depth-rectified statistical scattering modeling for deep-sea video descattering[J]. Infrared and Laser Engineering, 2022, 51(9): 20210919

    Download Citation

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

    Category: Image processing

    Received: Nov. 29, 2021

    Accepted: Jan. 13, 2022

    Published Online: Jan. 6, 2023

    The Author Email:

    DOI:10.3788/IRLA20210919

    Topics