Optics and Precision Engineering, Volume. 32, Issue 20, 3047(2024)

End-to-end deblurring model for microscopic vision

Zheng XU... Jiaheng HE, Yanqi WANG, Xiaodong WANG* and Tongqun REN |Show fewer author(s)
Author Affiliations
  • College of Mechanical Engineering, Dalian University of Technology, Dalian116081, China
  • show less
    Figures & Tables(18)
    Workflow of end-to-end deblurring model
    Process workflow of frequency domain
    Architecture of the blur discriminator.
    Architecture of deblurring network
    Comparison between gaussian blur kernel size and optical defocus distance
    Images with different blurriness degrees
    Overall training process
    Comparison of restoration results
    Recovery of image with multiple blurriness
    Experimental parts for precision measurement
    • Table 1. Blurriness discrimination results

      View table
      View in Article

      Table 1. Blurriness discrimination results

      类别精确率召回率F1值准确率
      清晰0.930.990.960.94
      轻度模糊0.990.990.99
      中度模糊0.880.970.92
      重度模糊0.970.840.90
    • Table 2. Frequency domain processing module ablation experiment

      View table
      View in Article

      Table 2. Frequency domain processing module ablation experiment

      组成部分选择
      傅里叶变换+极坐标转换
      低频剔除
      归一化

      清晰图片

      子数据集

      精确率0.931.000.980.990.77
      召回率0.990.990.971.001.00
      F1值0.961.000.970.990.87

      轻度模糊

      子数据集

      精确率1.000.970.990.900.51
      召回率0.990.900.960.940.44
      F1值0.990.930.980.920.48

      中度模糊

      子数据集

      精确率0.880.820.620.810.57
      召回率0.970.840.990.830.84
      F1值0.920.830.760.820.68

      重度模糊

      子数据集

      精确率0.970.850.970.910.96
      召回率0.840.890.730.850.96
      F1值0.900.870.830.880.96
      总数据集准确率0.940.910.890.900.69
    • Table 3. Recovery result of corresponding branch networks for different degrees of blurriness

      View table
      View in Article

      Table 3. Recovery result of corresponding branch networks for different degrees of blurriness

      类别SSIMPSNR
      恢复模糊恢复模糊
      轻度模糊0.9830.96238.734.8
      中度模糊0.9730.81037.227.8
      重度模糊0.7710.48227.622.1
      均值0.9090.75134.528.2
    • Table 4. SSIM scores of end-to-end deblurring model

      View table
      View in Article

      Table 4. SSIM scores of end-to-end deblurring model

      低度模糊部分中度模糊部分重度模糊部分整体图像
      多重模糊0.9600.7890.4040.769
      图像恢复0.9630.8630.7060.864
    • Table 5. Recovery result on fine-tuned datasets

      View table
      View in Article

      Table 5. Recovery result on fine-tuned datasets

      数据集类别图像SSIM
      σ多重模糊0.769
      图像恢复0.864
      σ0.25多重模糊0.792
      图像恢复0.852
      σ0.35多重模糊0.749
      图像恢复0.780
    • Table 6. Comparison experiment of discriminator

      View table
      View in Article

      Table 6. Comparison experiment of discriminator

      精确率召回率F1
      本文分类器0.8770.8730.874
      ResNet180.8550.8470.849
      ResNet500.8700.8630.865
      VGG160.9280.9290.928
    • Table 7. Comparison experiment of deblurring networks

      View table
      View in Article

      Table 7. Comparison experiment of deblurring networks

      SSIMPSNR
      本文网络0.52023.1
      SRGAN70.24516.6
      SRResNet70.41520.5
      VDSR220.29117.6
      SRCNN230.32520.2
    • Table 8. Comparison experiment of detection accuracy

      View table
      View in Article

      Table 8. Comparison experiment of detection accuracy

      基准距离平均误差最大误差
      轻度模糊组模糊847.466.669.42
      恢复5.498.55
      中度模糊组模糊12.4512.69
      恢复3.625.44
      重度模糊组模糊12.4316.02
      恢复3.494.18
    Tools

    Get Citation

    Copy Citation Text

    Zheng XU, Jiaheng HE, Yanqi WANG, Xiaodong WANG, Tongqun REN. End-to-end deblurring model for microscopic vision[J]. Optics and Precision Engineering, 2024, 32(20): 3047

    Download Citation

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

    Category:

    Received: Mar. 25, 2024

    Accepted: --

    Published Online: Jan. 10, 2025

    The Author Email: WANG Xiaodong (xdwang@dlut.edu.cn)

    DOI:10.37188/OPE.20243220.3047

    Topics