• Optics and Precision Engineering
  • Vol. 31, Issue 18, 2713 (2023)
Chengxi WANG1, Chen LUO2,*, Jianghao ZHOU2, Lang ZOU2, and Lei JIA3
Author Affiliations
  • 1School of Software Engineering, Southeast University, Suzhou2523, China
  • 2School of Mechanical Engineering, Southeast University, Nanjin11189, China
  • 3Wuxi shangshi-finevision Technology Co., Ltd, Wuxi200240, China
  • show less
    DOI: 10.37188/OPE.20233118.2713 Cite this Article
    Chengxi WANG, Chen LUO, Jianghao ZHOU, Lang ZOU, Lei JIA. Uniform defocus blind deblurring based on deeper feature-based wiener deconvolution[J]. Optics and Precision Engineering, 2023, 31(18): 2713 Copy Citation Text show less
    Overview architecture of uniform defocusing deblurring network
    Fig. 1. Overview architecture of uniform defocusing deblurring network
    Architecture of blur kernel estimation network
    Fig. 2. Architecture of blur kernel estimation network
    Architecture of deeper feature-based wiener deconvolution module
    Fig. 3. Architecture of deeper feature-based wiener deconvolution module
    Results of the non-blind deblurring algorithm FTVD
    Fig. 4. Results of the non-blind deblurring algorithm FTVD
    Comparison of PSNR under different blurring degrees
    Fig. 5. Comparison of PSNR under different blurring degrees
    Visual results of a test image in DIV2K
    Fig. 6. Visual results of a test image in DIV2K
    Visual results of a test image in GOPRO
    Fig. 7. Visual results of a test image in GOPRO
    Visual results of a true blurred image
    Fig. 8. Visual results of a true blurred image
    Visual results of conductive particle detection after blurring the true blurred image
    Fig. 9. Visual results of conductive particle detection after blurring the true blurred image
    MethodDIV2KGOPRO
    MSE/10-3PSNR/dBSSIMTimeMSE/10-3PSNR/dBSSIMTime
    SelfBlur205.74122.410.643 71.5 h4.63423.340.680 91.2 h
    DeepDeblur151.43928.420.798 40.04 s0.75331.230.895 50.03 s
    NAFNet171.29128.890.889 70.32 s0.75231.240.939 40.10 s
    SRN161.18329.270.839 70.05 s0.62532.040.913 10.05 s
    FWD231.11929.510.923 61.13 s0.45733.340.957 30.78 s
    UDBD(ours)0.76631.160.940 60.35 s0.24236.160.976 40.23 s
    Table 1. Comparison of objective performance indicators of different algorithms
    DFWDFRMGOPRO
    PSNR/dBSSIM
    ×30.330.892 6
    ×34.890.969 4
    36.160.976 4
    Table 2. Results of ablation experiment
    Chengxi WANG, Chen LUO, Jianghao ZHOU, Lang ZOU, Lei JIA. Uniform defocus blind deblurring based on deeper feature-based wiener deconvolution[J]. Optics and Precision Engineering, 2023, 31(18): 2713
    Download Citation