Optics and Precision Engineering, Volume. 33, Issue 1, 107(2025)
Full-reference image quality assessment based on texture singular value decomposition
Figures & Tables(16)
Fig. 1. Image summation of different subspace bases with 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
Fig. 3. Reference image
Fig. 5. Reference image
Fig. 6. First ten singular values of five contrast-change-distorted images
Fig. 8. Mean SROCC of 6 datasets for different
Fig. 9. Mean SROCC values of different
Fig. 10. Scatter plots and variances of MOS and objective evaluation scores on TID2013 and LIVE datasets
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）
Table 1. IQA databases
View tableView in ArticleTable 1. IQA databases
Database Distortion types Distortedimages Observers TID2013 24 3 000 971 TID2008 17 1 700 838 CSIQ 6 866 35 LIVE 5 779 161 CCID2014 1 655 22 IVC 4 185 15
Table 2. Comparison of overall performance of different FR-IQA on 6 data databases
View tableView in ArticleTable 2. Comparison of overall performance of different FR-IQA on 6 data databases
Method PLCC TID2 013
3 000
TID2 008
1 700
CCID2 014
655
CSIQ866 IVC185 LIVE779 加权平均 PSNR 0.492 0 0.502 8 0.174 3 0.878 6 0.719 3 0.872 1 0.746 5 SSIM 0.691 1 0.505 7 0.825 6 0.745 0 0.911 2 0.944 5 0.859 1 IWSSIM 0.831 1 0.867 1 0.860 2 0.914 2 0.923 1 0.962 1 0.875 2 FSIM 0.859 6 0.645 8 0.818 3 0.919 2 0.937 5 0.968 2 0.769 1 VIF 0.773 2 0.808 2 0.820 6 0.927 3 0.902 1 0.966 5 0.875 2 IFC 0.553 3 0.734 1 0.780 6 0.838 4 0.909 1 0.926 5 0.802 4 GMSD 0.859 3 0.878 2 0.807 3 0.954 2 0.923 4 0.960 2 0.890 4 SSVD 0.872 0 0.894 0 0.836 0 0.888 0 0.935 0 0.956 0 0.886 5 LPIPS 0.989 0 0.949 5 0.865 2 0.896 0 0.900 1 0.934 5 0.938 1 CPCC 0.878 1 0.886 8 0.867 4 0.953 7 0.923 2 0.961 6 0.898 1 TSVD 0.897 7 0.894 1 0.855 9 0.937 4 0.922 1 0.967 0 0.905 9 Method SROCC TID2 013
3 000
TID2 008
1 700
CCID2 014
655
CSIQ866 IVC185 LIVE779 加权平均 PSNR 0.523 0 0.546 8 0.690 8 0.805 8 0.688 4 0.875 6 0.688 7 SSIM 0.637 0 0.487 7 0.813 6 0.875 6 0.901 8 0.947 9 0.845 5 IWSSIM 0.777 1 0.655 1 0.724 4 0.921 3 0.912 5 0.956 7 0.811 2 FSIM 0.800 7 0.438 8 0.765 4 0.931 0 0.926 4 0.976 0 0.662 0 VIF 0.677 0 0.749 2 0.773 4 0.919 2 0.896 1 0.963 2 0.792 1 IFC 0.538 2 0.567 8 0.724 4 0.767 2 0.899 2 0.925 6 0.767 3 GMSD 0.804 2 0.890 1 0.768 8 0.953 3 0.914 3 0.964 1 0.877 1 SSVD 0.811 0 0.910 2 0.901 2 0.942 5 0.929 0 0.971 5 0.872 6 LPIPS 0.980 0 0.911 5 0.892 5 0.876 0 0.929 7 0.932 0 0.928 3 CPCC 0.859 2 0.890 8 0.768 1 0.954 2 0.925 4 0.977 1 0.886 1 TSVD 0.894 9 0.891 0 0.745 8 0.966 5 0.939 0 0.960 7 0.897 2 Method KROCC TID2 013
3 000
TID2 008
1 700
CCID2 014
655
CSIQ866 IVC185 LIVE779 加权平均 PSNR 0.333 1 0.375 1 0.492 8 0.692 4 0.521 8 0.686 5 0.517 2 SSIM 0.463 6 0.640 2 0.600 3 0.532 3 0.722 3 0.796 3 0.661 5 IWSSIM 0.597 2 0.695 3 0.535 6 0.752 9 0.733 9 0.817 5 0.638 2 FSIM 0.630 0 0.333 1 0.570 5 0.756 1 0.756 3 0.836 3 0.508 7 VIF 0.514 8 0.586 1 0.573 5 0.753 4 0.715 3 0.828 6 0.638 4 IFC 0.393 1 0.523 1 0.535 6 0.589 3 0.720 1 0.757 1 0.587 2 GMSD 0.633 1 0.727 0 0.571 1 0.802 3 0.756 8 0.826 8 0.696 3 SSVD 0.647 0 0.738 0 0.601 3 0.726 0 0.761 0 0.829 0 0.695 5 LPIPS 0.896 0 0.852 5 0.612 3 0.689 0 0.747 0 0.765 0 0.765 9 CPCC 0.673 8 0.704 0 0.605 5 0.800 9 0.815 0 0.835 6 0.711 2 TSVD 0.706 7 0.739 0 0.602 6 0.806 0 0.825 0 0.820 1 0.732 1
Table 3. Comparison of SROCC of IQA model for each distortion type in TID2013，LIVE and CSIQ datasets
View tableView in ArticleTable 3. Comparison of SROCC of IQA model for each distortion type in TID2013，LIVE and CSIQ datasets
Dataset Modal PSNR SSIM IWSSIM FSIM VIF IFC GMSD SSVD CPCC TSVD TID2013 AGN 0.929 1 0.864 6 0.843 8 0.897 3 0.899 4 0.661 2 0.922 1 0.946 2 0.932 5 0.932 6 ANC 0.898 1 0.773 0 0.751 5 0.820 8 0.829 9 0.535 2 0.868 4 0.828 9 0.843 4 0.828 7 SCN 0.920 0 0.854 4 0.816 7 0.875 0 0.883 5 0.660 1 0.935 0 0.936 9 0.911 9 0.931 5 MN 0.832 3 0.807 3 0.802 1 0.794 4 0.845 1 0.692 7 0.707 5 0.732 5 0.781 5 0.731 4 HFN 0.914 1 0.860 4 0.855 3 0.898 4 0.897 2 0.740 6 0.916 2 0.899 5 0.908 0 0.910 2 IN 0.796 8 0.762 9 0.728 2 0.807 2 0.854 3 0.640 8 0.763 7 0.767 1 0.781 4 0.767 1 QN 0.880 8 0.870 6 0.846 8 0.871 9 0.785 4 0.628 2 0.904 9 0.857 2 0.899 3 0.899 6 GB 0.914 6 0.967 3 0.970 1 0.955 1 0.965 1 0.890 7 0.911 3 0.949 9 0.926 8 0.986 2 DEN 0.948 0 0.926 8 0.915 2 0.930 2 0.891 1 0.778 5 0.952 0 0.948 6 0.946 6 0.968 2 JP2K 0.918 9 0.926 5 0.918 4 0.932 4 0.919 2 0.832 1 0.950 7 0.931 8 0.946 3 0.941 1 JP2K 0.884 0 0.950 4 0.950 6 0.957 7 0.951 6 0.807 8 0.968 8 0.965 7 0.954 7 0.852 7 JGTE 0.768 5 0.847 5 0.838 8 0.846 4 0.840 9 0.742 5 0.840 3 0.844 1 0.883 2 0.802 5 J2TE 0.888 3 0.888 9 0.865 6 0.891 3 0.876 1 0.776 9 0.913 6 0.933 2 0.880 5 0.911 5 NEPN 0.686 3 0.796 8 0.801 1 0.791 7 0.772 1 0.573 7 0.814 2 0.808 5 0.831 8 0.810 2 BLOCK 0.755 2 0.481 5 0.371 7 0.548 9 0.530 6 0.241 5 0.662 5 0.476 4 0.662 0 0.476 8 MS 0.767 1 0.790 7 0.783 3 0.753 1 0.627 6 0.552 1 0.795 1 0.750 7 0.710 7 0.808 2 CTC 0.441 0 0.463 5 0.459 3 0.468 6 0.638 7 0.179 8 0.323 5 0.449 4 0.314 7 0.859 5 CCS 0.476 6 0.409 9 0.419 6 0.474 8 0.309 9 0.402 9 0.394 8 0.356 2 0.584 0 0.401 5 MGN 0.890 5 0.778 6 0.772 8 0.846 5 0.846 2 0.614 3 0.888 6 0.860 0 0.868 8 0.939 2 CN 0.841 1 0.852 8 0.876 2 0.912 2 0.894 6 0.716 0.926 3 0.929 4 0.934 5 0.915 2 LCNI 0.914 5 0.906 8 0.903 5 0.946 6 0.920 4 0.818 1 0.962 9 0.964 9 0.965 6 0.963 4 ICQD 0.926 3 0.855 5 0.840 1 0.876 0 0.841 5 0.600 6 0.890 2 0.894 8 0.906 1 0.894 8 CHA 0.887 2 0.878 4 0.868 2 0.871 2 0.884 7 0.821 2 0.853 1 0.891 6 0.857 7 0.882 1 SSR 0.904 2 0.948 3 0.947 4 0.956 5 0.935 3 0.888 5 0.968 3 0.970 0 0.960 3 0.970 1 LIVE JPEG2K 0.894 3 0.961 4 0.964 9 0.971 7 0.969 6 0.911 3 0.971 1 0.969 9 0.970 7 0.947 8 JPEG 0.882 6 0.976 4 0.980 8 0.984 5 0.983 4 0.946 5 0.978 1 0.984 8 0.980 4 0.980 5 AWGN 0.974 5 0.969 4 0.966 7 0.985 8 0.965 3 0.938 2 0.973 7 0.980 1 0.981 4 0.966 9 GB 0.788 2 0.951 4 0.982 1 0.972 8 0.970 5 0.958 4 0.956 7 0.968 2 0.985 1 0.968 9 FF 0.889 6 0.955 2 0.944 2 0.965 2 0.948 9 0.962 3 0.941 6 0.954 6 0.968 0 0.939 5 CSIQ AGWN 0.936 3 0.897 4 0.938 0 0.935 9 0.957 4 0.843 2 0.967 5 0.947 4 0.949 0 0.952 4 JPEG 0.888 1 0.954 6 0.966 2 0.964 4 0.970 4 0.841 5 0.965 3 0.969 3 0.968 2 0.956 3 JP2K 0.936 2 0.960 6 0.968 2 0.970 4 0.967 2 0.925 2 0.971 8 0.969 2 0.973 0 0.979 5 AGPN 0.933 9 0.892 2 0.905 9 0.937 0 0.951 2 0.826 1 0.950 3 0.939 2 0.932 6 0.959 5 GB 0.929 2 0.960 6 0.978 2 0.972 4 0.974 5 0.952 4 0.961 2 0.971 3 0.975 4 0.986 5 GCD 0.862 4 0.792 2 0.933 6 0.943 8 0.934 1 0.487 3 0.903 9 0.869 0 0.927 4 0.965 2
Table 4. Comparison of operation speed of different FR-IQA methods
View tableView in ArticleTable 4. Comparison of operation speed of different FR-IQA methods
Method FPS PSNR 95.22 SSIM 30.75 IWSSIM 2.48 FSIM 3.58 VIF 1.56 IFC 4.58 GMSD 9.82 SSVD 20.55 LPIPS 105.38 CPCC 13.59 TSVD 22.05
Tools
Get Citation
Copy Citation Text
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
Paper Information
Category:
Received: Aug. 13, 2024
Accepted: --
Published Online: Apr. 1, 2025
The Author Email: Fajie DUAN (fjduan@tju.edu.cn)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20