IEEE Computational Intelligence Magazine - February 2023 - 78

TABLE II Average evaluation metrics of fused images on different datasets for different methods.
INPUTS METRIC IFCNN DDCGAN U2Fusion PMGI
FSIM "
UIQI "
QAB/F "
NMI "
ZL
Set-1
0.7563 0.6369
0.7165
0.0596
0.5135 0.2505
0.6662
0.5039
Set-2
FSIM "
UIQI "
QAB/F "
NMI "
Set-3
FSIM "
UIQI "
QAB/F "
NMI "
Set-4
FSIM "
UIQI "
QAB/F "
NMI "
Set-5
FSIM "
UIQI "
QAB/F "
NMI "
Set-6
FSIM "
UIQI "
QAB/F "
NMI "
Set-7
FSIM "
UIQI "
QAB/F "
NMI "
Set-8
FSIM "
UIQI "
QAB/F "
NMI "
0.5849 0.6517
0.2736 0.0522
0.0928 0.2606
0.4050 0.4792
0.5912 0.7214
0.3736 0.0635
0.1060 0.2657
0.4089 0.5303
0.5981 0.7049
0.3707 0.0851
0.1278 0.2676
0.3985 0.5028
0.5906 0.7388
0.3919 0.0512
0.0800 0.2527
0.3997 0.5487
0.7480 0.6070
0.6686
0.0123
0.5974 0.1899
0.6939 0.4320
0.8433 0.6976
0.5868
0.0045
0.5850 0.1314
0.7422 0.4856
0.8060
0.5951
0.5241 0.0008
0.6541 0.1486
0.6625 0.3174
0.7392 0.7383 0.7298 0.7441
0.4592 0.4881 0.5232 0.4130
0.7390 0.7278 0.7352 0.7598
0.5465
0.2677 0.1993 0.4659
0.5789
0.4748 0.4849
0.6305 0.6521 0.6083
0.7756
LRD CSMCA MSMG PAPCNN EMFusion OURS
0.7686
0.7355 0.7354 0.7846
0.2900 0.2282 0.5567 0.6386 0.6015 0.4491 0.4620 0.6934
0.5542
0.5120 0.5135 0.6841
0.6053 0.6619 0.6101 0.7030 0.6364 0.6044 0.6003 0.7791
0.7541 0.7322 0.7557
0.5247 0.4280 0.4233 0.4461
0.4771 0.5504 0.5336 0.5142 0.2882
0.7025
0.6592 0.6430 0.5930 0.7761
0.7995 0.7601 0.8208 0.8198
0.4494 0.5931 0.5878 0.5191
0.7310
0.6450 0.6391 0.6895
0.4028 0.2843 0.7693
0.7405 0.7114
0.8425
0.2738 0.2127 0.5733 0.6345 0.6406 0.4769 0.4472
0.6171
0.8654 0.7809 0.8813 0.8791 0.8831 0.8816
0.7775
0.5924 0.5685 0.6787 0.6416
0.8237
0.8456 0.7531 0.8615 0.8566
0.4673 0.5163 0.6075 0.5411
0.7016 0.6560 0.7625 0.7644
0.8287 0.8256 0.8399
0.6697
0.5665 0.5539 0.7138
0.7188 0.6485 0.6092 0.7773
0.8838
0.8840
0.7310 0.6784 0.6874 0.8104
0.6895
0.6698 0.6548 0.7323
0.8028 0.8006 0.7750 0.7213 0.8666
0.8732
0.2809 0.1916 0.6553 0.6909 0.6981 0.5399 0.5335
0.6408
0.7839
0.7173 0.6546 0.7410 0.7242 0.7423 0.7403
0.4107 0.1946 0.6803 0.5748 0.6685 0.6158
0.6426 0.6237 0.6009 0.6028 0.6699 0.6185
0.8380 0.7403 0.8496 0.8456
0.4952 0.1674 0.6656 0.6587
0.6776 0.6723 0.7433 0.7539
0.8529
0.8618 0.8581 0.8642
0.7302
0.5908 0.5564 0.7352
0.7043 0.6601 0.8117
0.7514
0.7307
0.2537 0.1338 0.5210 0.5800 0.6552 0.4547 0.6641 0.6062
0.6809
0.6973
0.6824
0.7182
0.8492 0.8498 0.8512
0.2323 0.0875 0.5544 0.5843 0.5709 0.5203 0.5749 0.5934
0.6969
0.7677
0.6911 0.6825 0.6068
0.7570 0.7579 0.7845
0.7855 0.6865 0.8002 0.7904 0.7952 0.7955 0.8033 0.7974
0.5529
0.7455
0.7819
0.1824 0.0779 0.4895 0.5301 0.5326 0.4792
0.4540 0.1393 0.7164 0.6396 0.7186 0.7062
0.5685 0.4600 0.6448 0.6301 0.6182 0.6737
The average best and second results are shown in RED and Blue BOLD font.
TABLE III Average metrics of the proposed method with
different regularization parameters in Lie.
PARAMETER
¼ 0:3
¼ 0:6
¼ 0:7
¼ 0:8
¼ 1:2
FSIM "
0.8232
0.8239
0.8272
0.8242
0.8234
UIQI "
0.6736
0.6700
0.6870
0.6689
0.6660
QAB/F "
0.8123
0.7976
0.7874
0.7949
0.7956
NMIN "
1.1194
1.1209
1.1247
1.1213
1.1134
The average best and second results are shown in RED and Blue BOLD font.
1)Ablation to Architecture
To verify the superiority of the network architecture for the
proposed method, we let the SSN-CAEþIE be degraded into
four versions: No SS (Self-Supervision), No Vit, No CNN,
and No IE. As shown in Fig. 12, the version without CNN
lost a large number oflocal features, whereas the version without
Vit lacked global features of source images. Therefore, IE
block is employed to better complete the fusion contribution
estimation of source images. From Table IV, it can observe
that SSN-CAEþIE has improved by 0.0317, 0.0638, 0.1569,
78 IEEE COMPUTATIONAL INTELLIGENCE MAGAZINE | FEBRUARY 2023
0.4987
0.6133
0.6046
0.7937
0.6939
0.7373
1.0977
0.7861
0.6099
0.7373
1.1425
0.8452
0.6680
0.7899
1.1229
0.8878
0.8156
0.8195
1.1633
0.8729
0.7251
0.7998
1.1559
0.7518
0.6795
0.8615
1.0699
0.8531
0.6168
0.7664
1.1209
0.8088
0.5694
0.8329
1.0903
IEEE Computational Intelligence Magazine - February 2023

Table of Contents for the Digital Edition of IEEE Computational Intelligence Magazine - February 2023
