• Optics and Precision Engineering
  • Vol. 33, Issue 1, 107 (2025)
Jiaxin LI, Fajie DUAN*, Xiao FU, and Guangyue NIU
Author Affiliations
  • State Key Laboratory of Precision Measuring Technology & Instruments, Tianjin University, Tianjin300072, China
  • show less
    DOI: 10.37188/OPE.20253301.0107 Cite this Article
    Jiaxin LI, Fajie DUAN, Xiao FU, Guangyue NIU. Full-reference image quality assessment based on texture singular value decomposition[J]. Optics and Precision Engineering, 2025, 33(1): 107 Copy Citation Text show less
    Image summation of different subspace bases with singular value weighting for images of size 384×512
    Fig. 1. Image summation of different subspace bases with singular value weighting for images of size 384×512
    Image summation of different subspace bases without singular value weighting for images of size 384×512
    Fig. 2. Image summation of different subspace bases without singular value weighting for images of size 384×512
    Reference image A (B4) and five images with average offset distortion
    Fig. 3. Reference image AB4) and five images with average offset distortion
    Top ten singular values of six images
    Fig. 4. Top ten singular values of six images
    Reference image A and 5 contrast-changing distorted images with increasing contrast
    Fig. 5. Reference image A and 5 contrast-changing distorted images with increasing contrast
    First ten singular values of five contrast-change-distorted images
    Fig. 6. First ten singular values of five contrast-change-distorted images
    Flowchart of proposed model based on texture SVD
    Fig. 7. Flowchart of proposed model based on texture SVD
    Mean SROCC of 6 datasets for different Thr and block sizes
    Fig. 8. Mean SROCC of 6 datasets for different Thr and block sizes
    Mean SROCC values of different γ and Tu block values in 6 datasets
    Fig. 9. Mean SROCC values of different γ and Tu block values in 6 datasets
    Scatter plots and variances of MOS and objective evaluation scores on TID2013 and LIVE datasets
    Fig. 10. Scatter plots and variances of MOS and objective evaluation scores on TID2013 and LIVE datasets
    Performance histogram of different methods
    Fig. 11. Performance histogram of different methods
    Statistical significance test results of IQA method('1' indicates that the method in the row is significantly better than the methods in the column, while '0' indicates that the method in the row is not significantly better than the method in the column)
    Fig. 12. Statistical significance test results of IQA method('1' indicates that the method in the row is significantly better than the methods in the column, while '0' indicates that the method in the row is not significantly better than the method in the column)
    DatabaseDistortion typesDistortedimagesObservers
    TID2013243 000971
    TID2008171 700838
    CSIQ686635
    LIVE5779161
    CCID2014165522
    IVC418515
    Table 1. IQA databases
    MethodPLCC

    TID2 013

    3 000

    TID2 008

    1 700

    CCID2 014

    655

    CSIQ866IVC185LIVE779加权平均
    PSNR0.492 00.502 80.174 30.878 60.719 30.872 10.746 5
    SSIM0.691 10.505 70.825 60.745 00.911 20.944 50.859 1
    IWSSIM0.831 10.867 10.860 20.914 20.923 10.962 10.875 2
    FSIM0.859 60.645 80.818 30.919 20.937 50.968 20.769 1
    VIF0.773 20.808 20.820 60.927 30.902 10.966 50.875 2
    IFC0.553 30.734 10.780 60.838 40.909 10.926 50.802 4
    GMSD0.859 30.878 20.807 30.954 20.923 40.960 20.890 4
    SSVD0.872 00.894 00.836 00.888 00.935 00.956 00.886 5
    LPIPS0.989 00.949 50.865 20.896 00.900 10.934 50.938 1
    CPCC0.878 10.886 80.867 40.953 70.923 20.961 60.898 1
    TSVD0.897 70.894 10.855 90.937 40.922 10.967 00.905 9
    MethodSROCC

    TID2 013

    3 000

    TID2 008

    1 700

    CCID2 014

    655

    CSIQ866IVC185LIVE779加权平均
    PSNR0.523 00.546 80.690 80.805 80.688 40.875 60.688 7
    SSIM0.637 00.487 70.813 60.875 60.901 80.947 90.845 5
    IWSSIM0.777 10.655 10.724 40.921 30.912 50.956 70.811 2
    FSIM0.800 70.438 80.765 40.931 00.926 40.976 00.662 0
    VIF0.677 00.749 20.773 40.919 20.896 10.963 20.792 1
    IFC0.538 20.567 80.724 40.767 20.899 20.925 60.767 3
    GMSD0.804 20.890 10.768 80.953 30.914 30.964 10.877 1
    SSVD0.811 00.910 20.901 20.942 50.929 00.971 50.872 6
    LPIPS0.980 00.911 50.892 50.876 00.929 70.932 00.928 3
    CPCC0.859 20.890 80.768 10.954 20.925 40.977 10.886 1
    TSVD0.894 90.891 00.745 80.966 50.939 00.960 70.897 2
    MethodKROCC

    TID2 013

    3 000

    TID2 008

    1 700

    CCID2 014

    655

    CSIQ866IVC185LIVE779加权平均
    PSNR0.333 10.375 10.492 80.692 40.521 80.686 50.517 2
    SSIM0.463 60.640 20.600 30.532 30.722 30.796 30.661 5
    IWSSIM0.597 20.695 30.535 60.752 90.733 90.817 50.638 2
    FSIM0.630 00.333 10.570 50.756 10.756 30.836 30.508 7
    VIF0.514 80.586 10.573 50.753 40.715 30.828 60.638 4
    IFC0.393 10.523 10.535 60.589 30.720 10.757 10.587 2
    GMSD0.633 10.727 00.571 10.802 30.756 80.826 80.696 3
    SSVD0.647 00.738 00.601 30.726 00.761 00.829 00.695 5
    LPIPS0.896 00.852 50.612 30.689 00.747 00.765 00.765 9
    CPCC0.673 80.704 00.605 50.800 90.815 00.835 60.711 2
    TSVD0.706 70.739 00.602 60.806 00.825 00.820 10.732 1
    Table 2. Comparison of overall performance of different FR-IQA on 6 data databases
    DatasetModalPSNRSSIMIWSSIMFSIMVIFIFCGMSDSSVDCPCCTSVD
    TID2013AGN0.929 10.864 60.843 80.897 30.899 40.661 20.922 10.946 20.932 50.932 6
    ANC0.898 10.773 00.751 50.820 80.829 90.535 20.868 40.828 90.843 40.828 7
    SCN0.920 00.854 40.816 70.875 00.883 50.660 10.935 00.936 90.911 90.931 5
    MN0.832 30.807 30.802 10.794 40.845 10.692 70.707 50.732 50.781 50.731 4
    HFN0.914 10.860 40.855 30.898 40.897 20.740 60.916 20.899 50.908 00.910 2
    IN0.796 80.762 90.728 20.807 20.854 30.640 80.763 70.767 10.781 40.767 1
    QN0.880 80.870 60.846 80.871 90.785 40.628 20.904 90.857 20.899 30.899 6
    GB0.914 60.967 30.970 10.955 10.965 10.890 70.911 30.949 90.926 80.986 2
    DEN0.948 00.926 80.915 20.930 20.891 10.778 50.952 00.948 60.946 60.968 2
    JP2K0.918 90.926 50.918 40.932 40.919 20.832 10.950 70.931 80.946 30.941 1
    JP2K0.884 00.950 40.950 60.957 70.951 60.807 80.968 80.965 70.954 70.852 7
    JGTE0.768 50.847 50.838 80.846 40.840 90.742 50.840 30.844 10.883 20.802 5
    J2TE0.888 30.888 90.865 60.891 30.876 10.776 90.913 60.933 20.880 50.911 5
    NEPN0.686 30.796 80.801 10.791 70.772 10.573 70.814 20.808 50.831 80.810 2
    BLOCK0.755 20.481 50.371 70.548 90.530 60.241 50.662 50.476 40.662 00.476 8
    MS0.767 10.790 70.783 30.753 10.627 60.552 10.795 10.750 70.710 70.808 2
    CTC0.441 00.463 50.459 30.468 60.638 70.179 80.323 50.449 40.314 70.859 5
    CCS0.476 60.409 90.419 60.474 80.309 90.402 90.394 80.356 20.584 00.401 5
    MGN0.890 50.778 60.772 80.846 50.846 20.614 30.888 60.860 00.868 80.939 2
    CN0.841 10.852 80.876 20.912 20.894 60.7160.926 30.929 40.934 50.915 2
    LCNI0.914 50.906 80.903 50.946 60.920 40.818 10.962 90.964 90.965 60.963 4
    ICQD0.926 30.855 50.840 10.876 00.841 50.600 60.890 20.894 80.906 10.894 8
    CHA0.887 20.878 40.868 20.871 20.884 70.821 20.853 10.891 60.857 70.882 1
    SSR0.904 20.948 30.947 40.956 50.935 30.888 50.968 30.970 00.960 30.970 1
    LIVEJPEG2K0.894 30.961 40.964 90.971 70.969 60.911 30.971 10.969 90.970 70.947 8
    JPEG0.882 60.976 40.980 80.984 50.983 40.946 50.978 10.984 80.980 40.980 5
    AWGN0.974 50.969 40.966 70.985 80.965 30.938 20.973 70.980 10.981 40.966 9
    GB0.788 20.951 40.982 10.972 80.970 50.958 40.956 70.968 20.985 10.968 9
    FF0.889 60.955 20.944 20.965 20.948 90.962 30.941 60.954 60.968 00.939 5
    CSIQAGWN0.936 30.897 40.938 00.935 90.957 40.843 20.967 50.947 40.949 00.952 4
    JPEG0.888 10.954 60.966 20.964 40.970 40.841 50.965 30.969 30.968 20.956 3
    JP2K0.936 20.960 60.968 20.970 40.967 20.925 20.971 80.969 20.973 00.979 5
    AGPN0.933 90.892 20.905 90.937 00.951 20.826 10.950 30.939 20.932 60.959 5
    GB0.929 20.960 60.978 20.972 40.974 50.952 40.961 20.971 30.975 40.986 5
    GCD0.862 40.792 20.933 60.943 80.934 10.487 30.903 90.869 00.927 40.965 2
    Table 3. Comparison of SROCC of IQA model for each distortion type in TID2013,LIVE and CSIQ datasets
    MethodFPS
    PSNR95.22
    SSIM30.75
    IWSSIM2.48
    FSIM3.58
    VIF1.56
    IFC4.58
    GMSD9.82
    SSVD20.55
    LPIPS105.38
    CPCC13.59
    TSVD22.05
    Table 4. Comparison of operation speed of different FR-IQA methods
    Jiaxin LI, Fajie DUAN, Xiao FU, Guangyue NIU. Full-reference image quality assessment based on texture singular value decomposition[J]. Optics and Precision Engineering, 2025, 33(1): 107
    Download Citation