Laser & Optoelectronics Progress, Volume. 60, Issue 6, 0610002(2023)

Digital Generation Technology for Tomb Murals Based on Multiscale Cascade Network

Meng Wu1,*... Yi Ren1 and Jia Wang2 |Show fewer author(s)
Author Affiliations
  • 1School of Information and Control Engineering, Xi'an University of Architecture and Technology, Xi'an 710055, Shaanxi, China
  • 2Shaanxi History Museum, Xi'an 710061, Shaanxi, China
  • show less
    Figures & Tables(17)
    Schematic diagram of generating countermeasure network
    Multi-scale cascade network framework
    Specific structure diagram of upsampling and downsampling
    Epitaxial large-scale generation network
    Deep semantic small-scale generation network
    Self attention mechanism
    Discriminant network model
    Comparison of polo murals before and after restoration. (a) Original mural; (b) restored mural
    Loss convergence curve
    Comparison diagram of channel characteristics
    Ablation experiment
    Comparison of generation results of different algorithms
    • Table 1. Model convergence under different values

      View table

      Table 1. Model convergence under different values

      χ1χ2Epoch
      0.70.3188
      0.50.5166
      0.40.6143
      0.30.7133
      0.20.8110
      0.10.9Non convergence
      0.0010.999Non convergence
    • Table 2. Analysis of generation results between different layers

      View table

      Table 2. Analysis of generation results between different layers

      Location of self attention mechanismSSIMPSNR /dB
      2-3-layer0.7431.02
      3-4-layer0.8431.86
      4-5-layer0.6920.09
    • Table 3. Network and loss group name

      View table

      Table 3. Network and loss group name

      NameNetwork
      ED_1ED+Lrec+Ladv
      ED_2ED+Lrec+Ladv+Ltexture-loss
      ED_3ED+Lrec+Ladv+Ltex
      ED_4ED+Lrec+Ladv+Ltex+self-attention
    • Table 4. Comparison of generated by different methods

      View table

      Table 4. Comparison of generated by different methods

      MethodWood blockMountain blockTomb passage block
      PSNR /dBSSIMPSNR /dBSSIMPSNR /dBSSIM
      ED_125.290.6927.610.6227.860.70
      ED_228.230.7329.690.7130.400.77
      ED_330.660.8031.740.8331.910.82
      ED_431.030.8532.310.8632.820.88
    • Table 5. Comparison of generation results of different algorithms

      View table

      Table 5. Comparison of generation results of different algorithms

      MethodHorse blockStone blockTomb passage blockWood block
      PSNR /dBSSIMPSNR /dBSSIMPSNR /dBSSIMPSNR /dBSSIM
      WGAN-GP27.350.6824.870.5726.360.6527.850.64
      DCGAN30.160.7829.690.7231.590.8029.470.72
      Proposed32.660.8431.740.8032.910.8831.660.84
    Tools

    Get Citation

    Copy Citation Text

    Meng Wu, Yi Ren, Jia Wang. Digital Generation Technology for Tomb Murals Based on Multiscale Cascade Network[J]. Laser & Optoelectronics Progress, 2023, 60(6): 0610002

    Download Citation

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

    Category: Image Processing

    Received: Nov. 23, 2021

    Accepted: Jan. 11, 2022

    Published Online: Mar. 7, 2023

    The Author Email: Meng Wu (wumeng@xauat.edu.cn)

    DOI:10.3788/LOP213032

    Topics