• Optics and Precision Engineering
  • Vol. 30, Issue 18, 2280 (2022)
Chunhua WU, Hong PENG, Qiegen LIU, Wenbo WAN*, and Yuhao WANG
Author Affiliations
  • School of Information Engineering, Nanchang University, Nanchang330031, China
  • show less
    DOI: 10.37188/OPE.20223018.2280 Cite this Article
    Chunhua WU, Hong PENG, Qiegen LIU, Wenbo WAN, Yuhao WANG. Lens-less imaging via score-based generative model[J]. Optics and Precision Engineering, 2022, 30(18): 2280 Copy Citation Text show less
    Flow chart of lens-less imaging
    Fig. 1. Flow chart of lens-less imaging
    Generative model
    Fig. 2. Generative model
    Forward and reverse time process of SDE
    Fig. 3. Forward and reverse time process of SDE
    Training and reconstruction flow chart of LSGM algorithm
    Fig. 4. Training and reconstruction flow chart of LSGM algorithm
    Visual comparison of reconstruction images on the LSUN-bedroom dataset
    Fig. 5. Visual comparison of reconstruction images on the LSUN-bedroom dataset
    Visual comparison of reconstruction images on the LSUN-church dataset
    Fig. 6. Visual comparison of reconstruction images on the LSUN-church dataset
    Improvement of visual effects by LSGM algorithm with TV method
    Fig. 7. Improvement of visual effects by LSGM algorithm with TV method
    Ablation experiments on the relationship between image quality and discretization steps under different σmax
    Fig. 8. Ablation experiments on the relationship between image quality and discretization steps under different σmax
    Experimental test of field of view
    Fig. 9. Experimental test of field of view
    Experimental test of the resolution
    Fig. 10. Experimental test of the resolution
    MethodPSNR/dBSSIM
    BP7.780.23
    CS18.470.31
    LSGM26.100.64
    Table 1. Test results on LSUN-bedroom dataset

    Algorithm 1 LSGM

    Training

    Dataset: Object image dataset

    Output: Sθ(O,t)Ologpt(O)

    Reconstruction

    Setting: Score network Sθ, Noise iteration step εi, Noise scale σi,Number of discretization steps for the reverse-time SDE N, Number of corrector steps M

    1: ONN(0,σmax2I)

    2: For i=N-1 to 0 do (Outer loop)

    3:  Update Oi via Eq. (12) and (13)   【Predictor】

    4:  For j=1 to M do (Inner loop)

    5:   Update Oi via Eq. (14) and (15) 【Corrector】

    6:  End for

    7: End for

    8: Return O0

    Table 1. [in Chinese]
    MethodPSNR/dBSSIM
    BP7.810.22
    CS17.790.29
    LSGM24.470.65
    Table 2. Test results on LSUN-church dataset
    Chunhua WU, Hong PENG, Qiegen LIU, Wenbo WAN, Yuhao WANG. Lens-less imaging via score-based generative model[J]. Optics and Precision Engineering, 2022, 30(18): 2280
    Download Citation