Chinese Journal of Liquid Crystals and Displays, Volume. 38, Issue 2, 245(2023)

Face image repair network based on face structure guidance

Hao-de SHI1, Ming-ju CHEN1,2、*, Jin HOU1,2, and Lan LI1
Author Affiliations
  • 1School of Automation and Information Engineering,Sichuan University of Science & Engineering,Yibin 644000,China
  • 2Artificial Intelligence Key Lab of Sichuan Province,Sichuan University of Science & Engineering,Yibin 644000,China
  • show less
    Figures & Tables(11)
    U-Net network structure diagram
    Diagram of GAN network structure
    Face structure sketch generation network frame diagram
    Frame diagram of face inpainting network
    Model repair test results in this paper.(a)Original image;(b)Sketch of real facial structure;(c)Occlusion of the face image;(d)Occlusion sketches of facial structures;(e)Generated structural sketch;(f)Repaired face image.
    Qualitative comparison of experimental results of random mask repair.(a)Original image;(b)Occluded image;(c)EC algorithm;(d)LaFIn algorithm;(e)CTSDG algorithm;(f)Our algorithm.
    Qualitative comparison of experimental results of center mask repair.(a)Original image;(b)Occluded image;(c)EC algorithm;(d)LaFIn algorithm;(e)CTSDG algorithm;(f)Our algorithm.
    Qualitative experimental repair details .(a)Original image;(b)EC algorithm;(c)LaFIn algorithm;(d)CTSDG algorithm;(e)Our algorithm.
    Qualitative analysis of ablation repair experiment.(a)Original image;(b)Occluded image;(c)Benchmark algorithm;(d)Sketch structure guidance;(e)Sketch structure guidance+attention mechanism.
    • Table 1. Quantitative comparison results of repair experiments under different masks

      View table
      View in Article

      Table 1. Quantitative comparison results of repair experiments under different masks

      评估指标掩膜类别ECLaFInCTSDG本文方法
      PSNR10%~20%27.3229.7829.9430.76
      20%~30%26.7227.4627.4828.09
      30%~40%23.1025.7726.3126.53
      40%~50%21.0723.4923.7824.07
      50%~60%18.7419.4320.1220.46
      中心掩膜21.5322.8922.0323.51
      SSIM10%~20%0.9260.9640.9760.978
      20%~30%0.8930.9210.9310.939
      30%~40%0.8410.8740.8810.895
      40%~50%0.8010.8520.8580.873
      50%~60%0.7380.7580.7690.772
      中心掩膜0.8170.8620.8390.896
      FID10%~20%3.683.233.173.04
      20%~30%4.974.654.514.10
      30%~40%9.497.747.467.13
      40%~50%14.989.689.529.01
      50%~60%22.4317.7217.0816.84
      中心掩膜13.299.048.768.21
    • Table 2. Quantitative comparison of results of ablation repair experiments

      View table
      View in Article

      Table 2. Quantitative comparison of results of ablation repair experiments

      评估指标测评对象基准SketchSketch+SA
      PSNR第①行23.0225.1225.44
      第②行23.8926.0326.76
      第③行25.1826.9427.32
      第④行24.3124.9225.69
      第⑤行23.9325.4226.11
      第⑥行22.9727.3227.71
      SSIM第①行0.8690.8990.904
      第②行0.9050.9250.939
      第③行0.8760.8950.915
      第④行0.8350.8490.855
      第⑤行0.8590.8830.897
      第⑥行0.8740.8800.896
      FID第①行25.317.218.92
      第②行39.521.3515.24
      第③行42.7119.2112.65
      第④行41.7423.5617.66
      第⑤行37.921.3415.9
      第⑥行38.723.6314.35
    Tools

    Get Citation

    Copy Citation Text

    Hao-de SHI, Ming-ju CHEN, Jin HOU, Lan LI. Face image repair network based on face structure guidance[J]. Chinese Journal of Liquid Crystals and Displays, 2023, 38(2): 245

    Download Citation

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

    Category: Research Articles

    Received: May. 28, 2022

    Accepted: --

    Published Online: Feb. 20, 2023

    The Author Email: Ming-ju CHEN (12347259@qq.com)

    DOI:10.37188/CJLCD.2022-0181

    Topics