• Laser & Optoelectronics Progress
  • Vol. 57, Issue 8, 081003 (2020)
Tingting Liu1, Yujin Zhang1、2、*, Fei Wu1, and Shiting Xiong1
Author Affiliations
  • 1School of Electrical and Electronic Engineering, Shanghai University of Engineering Science, Shanghai 201620, China
  • 2Shanghai Key Laboratory of Integrated Administration Technologies for Information Security, Shanghai Jiao Tong University, Shanghai 200240, China
  • show less
    DOI: 10.3788/LOP57.081003 Cite this Article Set citation alerts
    Tingting Liu, Yujin Zhang, Fei Wu, Shiting Xiong. Diffusion-Based Image Inpainting Forensics Via Gradient Domain Guided Filtering Enhancement[J]. Laser & Optoelectronics Progress, 2020, 57(8): 081003 Copy Citation Text show less
    Training and detection flow chart
    Fig. 1. Training and detection flow chart
    Artifacts left after the image has been inpainted. (a) Original image; (b) ground truth; (c) δΔI map
    Fig. 2. Artifacts left after the image has been inpainted. (a) Original image; (b) ground truth; (c) δΔI map
    Comparison of repair image positioning results. (a) Original image; (b) ground truth image; (c) location result obtained by algorithm used in reference [10]; (d) location result obtained by our algorithm
    Fig. 3. Comparison of repair image positioning results. (a) Original image; (b) ground truth image; (c) location result obtained by algorithm used in reference [10]; (d) location result obtained by our algorithm
    Pixel sizeAlgorithm
    SquareCircularIrregular
    Reference[10]ProposedReference[10]ProposedReference[10]Proposed
    64×640.88960.90880.88630.90560.87570.8947
    32×320.79080.81920.78930.81760.76450.7913
    16×160.65850.68800.66640.69600.62150.6489
    8×80.40630.43810.48580.51740.19900.2137
    Table 1. Comparison of F1 scores based on Isotropic repair algorithm
    Pixel sizeAlgorithm
    SquareCircularIrregular
    Reference[10]ProposedReference[10]ProposedReference[10]Proposed
    64×640.88210.90110.87900.89770.86610.8846
    32×320.78080.80820.77570.80310.74800.7735
    16×160.64650.67790.64420.67290.59400.6208
    8×80.31710.33880.40650.43150.17260.1833
    Table 2. Comparison of F1 score based on Edge-oriented repair algorithm
    Pixel sizeAlgorithm
    SquareCircularIrregular
    Reference[10]ProposedReference[10]ProposedReference[10]Proposed
    64×640.81470.83180.86100.87880.82970.8474
    32×320.61890.63880.69530.71640.67680.6977
    16×160.31380.33520.31840.33900.38330.4063
    8×80.10070.10740.12340.13050.07840.0784
    Table 3. Comparison of F1 scores based on Delaunay-oriented repair algorithm
    Pixel sizeAlgorithm
    SquareCircularIrregular
    Reference[10]ProposedReference[10]ProposedReference[10]Proposed
    64×640.81060.83360.80970.83350.79740.8194
    32×320.65140.68280.65260.68520.62890.6602
    16×160.45820.49150.46610.50070.43010.4626
    8×80.20510.22170.27420.29820.12460.1373
    Table 4. Comparison of F1 scores based on Isotropic repair algorithm (1.1×)
    Pixel sizeAlgorithm
    SquareCircularIrregular
    Reference[10]ProposedReference[10]ProposedReference[10]Proposed
    64×640.80010.82370.79810.82030.78500.8059
    32×320.63510.66590.63420.66630.60780.6366
    16×160.44470.47770.44270.47690.40790.4365
    8×80.15820.17290.22220.24370.09720.1085
    Table 5. Comparison of F1 score based on Edge-oriented repair algorithm (1.1×)
    Pixel sizeAlgorithm
    SquareCircularIrregular
    Reference[10]ProposedReference[10]ProposedReference[10]Proposed
    64×640.73740.75740.76870.78790.74660.7689
    32×320.50330.52560.54250.57140.53980.5653
    16×160.23110.24880.22800.24760.25880.2819
    8×80.04500.05170.06030.06530.04360.0479
    Table 6. Comparison of F1 scores based on Delaunay-oriented repair algorithm (1.1×)
    Tingting Liu, Yujin Zhang, Fei Wu, Shiting Xiong. Diffusion-Based Image Inpainting Forensics Via Gradient Domain Guided Filtering Enhancement[J]. Laser & Optoelectronics Progress, 2020, 57(8): 081003
    Download Citation