Matter and Radiation at Extremes, Volume. 10, Issue 2, 027402(2025)

Single-image super-resolution of gamma-ray imaging system using deep denoiser prior based on plug-and-play framework

Guo-Guang Li1,2,3、*, Liang Sheng4, Bao-Jun Duan4, Yang Li4, Yan Song4, Zi-Jian Zhu4, Wei-Peng Yan4, Dong-Wei Hei4, and Qing-Zi Xing1,2,3
Author Affiliations
  • 1Key Laboratory of Particle and Radiation Imaging (Tsinghua University), Ministry of Education, Beijing 100084, China
  • 2Laboratory for Advanced Radiation Sources and Application, Tsinghua University, Beijing 100084, China
  • 3Department of Engineering Physics, Tsinghua University, Beijing 100084, China
  • 4State Key Laboratory of Intense Pulsed Radiation Simulation and Effect, Northwest Institute of Nuclear Technology, Xi’an 710024, China
  • show less
    Figures & Tables(11)
    Schematic of gamma-ray imaging system. (b) Interaction process of lens-coupled scintillator. (c) Down-sampling process of CMOS camera.
    (a) Schematic illustration of the role of image boundaries in the convolution operation. (b) Illustration of the periodic boundary condition. (c) Illustration of the mask.
    Illustration of the contrast of nonuniform line-pairs.
    (a1)–(a3) Illustration of the imaging process considering the effects of boundaries, including blurring, downsampling, and noise. (b1) and (b2) Reconstruction results using PnP-BM3D with periodic boundaries. (c1) and (c2) Reconstruction results using PnP-DRUNet with periodic boundaries. (d1) and (d2) Reconstruction results using PnP-BM3D with estimated boundaries. (e1) and (e2) Reconstruction results using PnP-DRUNet with estimated boundaries. a.u., arbitrary units.
    (a) Original HR image containing rich shape information. (b1)–(b3) Recorded images with 1% Gaussian noise at s = 2, 3, and 4, respectively. (c1)–(c3) Reconstruction results using PnP-BM3D with estimated boundaries at s = 2, 3, and 4, respectively. (d1)–(d3) Reconstruction results using PnP-DRUNet with estimated boundaries at s = 2, 3, and 4, respectively.
    (a1) Original HR image containing four different widths (100, 150, 200, and 250 μm) of line-pairs. (a2) Schematic of subpixel offsets due to different starting positions for downsampling. (b1)–(b4), (c1)–(c4), (d1)–(d4), and (e1)–(e4) Images recorded under 0, 1/4, 2/4, and 3/4 subpixel offsets, and the corresponding reconstructed images using PnP-DRUNet with estimated boundaries at s = 2, 3, and 4, respectively. (b5)–(b8), (c5)–(c8), (d5)–(d8), and (e5)–(e8) Average intensity curves of line-pairs in (b1)–(b4), (c1)–(c4), (d1)–(d4), and (e1)–(e4), respectively. (f) Contrast of the 150 μm-wide line-pairs in the reconstruction results.
    (a) Photograph of the gamma-ray imaging system implemented in real experiments. (b) Photograph of the sample. (c) Raw gamma-ray radiation image attenuated by the sample. (d) Dark-field image from the gamma-ray imaging system. (e) Flat-field image from the gamma-ray imaging system. (f) Corrected gamma-ray radiation image attenuated by the sample. (g) Pseudocolor ROI gamma-ray radiation image attenuated by the sample. (h) Contrast of line-pairs within ROI image. (i) PSF curve of the gamma-ray imaging system.
    Reconstructed images using (a1)–(a3) PnP-BM3D under the periodic boundary assumption, (b1)–(b3) PnP-DRUNet under the periodic boundary assumption, (c1)–(c3) the proposed boundary estimation method with PnP-BM3D, and (d1)–(d3) the proposed boundary estimation method with PnP-DRUNet.
    Contrast of line-pairs with widths of (a) 250 μm, (b) 278 μm, and (c) 417 μm using different algorithms.
    • Table 1. Single image super-resolution using deep denoiser prior based on plug-and play framework (PnP-DRUNet).

      View table
      View in Article

      Table 1. Single image super-resolution using deep denoiser prior based on plug-and play framework (PnP-DRUNet).

    • Table 1. PSNR and SSIM results for reconstructed images with different boundaries and algorithms. Boldface denotes that the image assessment index of the corresponding algorithm is better than that of other algorithms.

      View table
      View in Article

      Table 1. PSNR and SSIM results for reconstructed images with different boundaries and algorithms. Boldface denotes that the image assessment index of the corresponding algorithm is better than that of other algorithms.

      BoundariesAlgorithmPSNRSSIM
      PeriodicPnP-BM3D21.290.7616
      PnP-DRUNet22.290.7951
      EstimatedPnP-BM3D25.230.8188
      PnP-DRUNet26.390.8501
    Tools

    Get Citation

    Copy Citation Text

    Guo-Guang Li, Liang Sheng, Bao-Jun Duan, Yang Li, Yan Song, Zi-Jian Zhu, Wei-Peng Yan, Dong-Wei Hei, Qing-Zi Xing. Single-image super-resolution of gamma-ray imaging system using deep denoiser prior based on plug-and-play framework[J]. Matter and Radiation at Extremes, 2025, 10(2): 027402

    Download Citation

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

    Received: Sep. 1, 2024

    Accepted: Dec. 18, 2024

    Published Online: Apr. 30, 2025

    The Author Email:

    DOI:10.1063/5.0236541

    Topics