Institute for Quantum Information & State Key Laboratory of High Performance Computing, College of Computer, National University of Defense Technology, Changsha 410073, China
In recent years, image fusion has emerged as an important research field due to its various applications. Images acquired by different sensors have significant differences in feature representation due to the different imaging principles. Taking visible and infrared image fusion as an example, visible images contain abundant texture details with high spatial resolution. In contrast, infrared images can obtain clear target contour information according to the principle of thermal radiation, and work well in all day/night and all weather conditions. Most existing methods employ the same feature extraction algorithm to get the feature information from visible and infrared images, ignoring the differences among these images. Thus, this paper proposes what we believe to be a novel fusion method based on a multi-level image decomposition method and deep learning fusion strategy for multi-type images. In image decomposition, we not only utilize a multi-level extended approximate low-rank projection matrix learning decomposition method to extract salient feature information from both visible and infrared images, but also apply a multi-level guide filter decomposition method to obtain texture information in visible images. In image fusion, a novel fusion strategy based on a pretrained ResNet50 network is presented to fuse multi-level feature information from both visible and infrared images into corresponding multi-level fused feature information, so as to improve the quality of the final fused image. The proposed method is evaluated subjectively and objectively in a large number of experiments. The experimental results demonstrate that the proposed method exhibits better fusion performance than other existing methods.
Test image data is available in Ref. [38]. Other data underlying the results presented in this paper are not publicly available at this time but may be obtained from the authors upon reasonable request.
You do not have subscription access to this journal. Cited by links are available to subscribers only. You may subscribe either as an Optica member, or as an authorized user of your institution.
You do not have subscription access to this journal. Figure files are available to subscribers only. You may subscribe either as an Optica member, or as an authorized user of your institution.
You do not have subscription access to this journal. Article tables are available to subscribers only. You may subscribe either as an Optica member, or as an authorized user of your institution.
You do not have subscription access to this journal. Equations are available to subscribers only. You may subscribe either as an Optica member, or as an authorized user of your institution.
Input: Input images ; number of decomposition level ; and parameters .
Initialization:; .
1:
2:
3: for to do
4:
5:
6: do EALPL decomposition algorithm, obtain
7: do visible feature fusion algorithm, obtain
8:
9: do feature fusion algorithm, obtain
10:
11: end for
12:
13:
Table 1.
Computing Equations of Quality Metrics Used in Our Experiments
Quality Metrics
Computing Equations
Table 2.
Fusion Performance of the Proposed Method with Different Ratios of Detail and Smooth Blocksa
Different Ratios of Detail and Smooth Blocks
Quality Metrics
[2000,0]
[1500, 500]
[1000, 1000]
[500, 1500]
[0, 2000]
6.7651
6.7713
6.7760
6.7770
6.7693
13.5302
13.5426
13.5520
13.5540
13.5387
0.4393
0.4400
0.4403
0.4410
0.4460
0.8997
0.8999
0.8997
0.8997
0.8998
0.4358
0.4358
0.4358
0.4359
0.4363
1.6946
1.6944
1.6948
1.6965
1.6993
0.7211
0.7283
0.7336
0.7344
0.7288
0.9113
0.9107
0.9103
0.9106
0.9137
The first value represents the number of detail blocks, and the second value denotes the number of smooth blocks. The average values of eight quality metrics for all test images are shown. The best scores are shown in boldface, and second-best values are marked in italic.
Table 3.
Fusion Performance of the Proposed Method with Different Weight Values of Base Imagesa
Weight Value of Base Images
Quality Metrics
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
6.9813
6.9142
6.8491
6.7997
6.7770
6.7817
6.8106
6.8619
6.9323
13.9625
13.8283
13.6981
13.5993
13.5540
13.5634
13.6212
13.7238
13.8645
0.4295
0.4333
0.4371
0.4398
0.4410
0.4406
0.4388
0.4356
0.4308
0.9002
0.9002
0.9002
0.9000
0.8997
0.8994
0.8989
0.8983
0.8976
0.3928
0.3947
0.3958
0.3961
0.3961
0.3957
0.3950
0.3941
0.3926
0.4351
0.4359
0.4362
0.4362
0.4359
0.4355
0.4349
0.4341
0.4331
0.4171
0.4193
0.4208
0.4221
0.4233
0.4245
0.4257
0.4270
0.4279
1.6468
1.6957
1.7117
1.7104
1.6965
1.6692
1.6217
1.5384
1.3853
0.6902
0.7075
0.7213
0.7302
0.7344
0.7337
0.7290
0.7208
0.7091
0.6455
0.6493
0.6516
0.6525
0.6520
0.6501
0.6468
0.6420
0.6356
0.9031
0.9075
0.9102
0.9113
0.9106
0.9083
0.9043
0.8986
0.8915
Average values of 11 quality metrics for all test images. The best scores are shown in boldface, second-best values are marked in italic, and third-best values are underlined.
Average values of eight quality metrics for all test images. The best scores are shown in boldface, and second-best values are marked in italic.
Table 6.
Average Values of Nine Quality Metrics for All Test Imagesa
Quality Metrics
Method
CBF
6.8575
13.7150
0.4396
0.8720
0.2631
0.3235
0.7145
1.3896
0.7088
JSR
6.7226
12.7265
0.3586
0.8846
0.1644
0.2083
0.6385
1.7507
0.7552
JSRSD
6.7206
13.3858
0.3261
0.8643
0.1444
0.1850
0.6707
1.5980
0.7552
GTF
6.6343
13.2687
0.4104
0.9021
0.4227
0.4104
0.4169
1.0134
0.8084
WLS
6.6379
13.2757
0.5008
0.8979
0.3349
0.3766
0.7287
1.7889
0.9335
ConvSR
6.2587
12.5174
0.5349
0.8902
0.1753
0.3842
0.3922
1.1490
0.9028
VggML
6.1826
12.3652
0.3677
0.9107
0.4050
0.4168
0.2951
1.6348
0.8748
ResNet-ZCA
6.1953
12.3905
0.3510
0.9092
0.4058
0.4169
0.2925
1.6336
0.8732
FusionGAN
6.3470
12.6941
0.1439
0.7963
0.1111
0.3708
0.4536
0.6185
0.7318
LatLRR
6.3574
12.7149
0.4128
0.8998
0.3382
0.3826
0.2974
1.7070
0.8757
MDLatLRR
6.9777
13.9555
0.3357
0.8955
0.3780
0.4448
0.8759
1.6328
0.8522
Ours
level-3
6.5776
13.1551
0.5042
0.9067
0.4074
0.4332
0.5521
1.6799
0.9399
level-4
6.7997
13.5993
0.4398
0.9000
0.3961
0.4362
0.7302
1.7104
0.9113
level-5
6.9864
13.9637
0.3306
0.8905
0.3816
0.4344
0.9039
1.6149
0.8461
The best scores are shown in boldface, second-best values are marked in in italic, and third-best values are underlined.
Tables (7)
Algorithm 1.
Visible and Infrared Image Fusion Method
Input: Input images ; number of decomposition level ; and parameters .
Initialization:; .
1:
2:
3: for to do
4:
5:
6: do EALPL decomposition algorithm, obtain
7: do visible feature fusion algorithm, obtain
8:
9: do feature fusion algorithm, obtain
10:
11: end for
12:
13:
Table 1.
Computing Equations of Quality Metrics Used in Our Experiments
Quality Metrics
Computing Equations
Table 2.
Fusion Performance of the Proposed Method with Different Ratios of Detail and Smooth Blocksa
Different Ratios of Detail and Smooth Blocks
Quality Metrics
[2000,0]
[1500, 500]
[1000, 1000]
[500, 1500]
[0, 2000]
6.7651
6.7713
6.7760
6.7770
6.7693
13.5302
13.5426
13.5520
13.5540
13.5387
0.4393
0.4400
0.4403
0.4410
0.4460
0.8997
0.8999
0.8997
0.8997
0.8998
0.4358
0.4358
0.4358
0.4359
0.4363
1.6946
1.6944
1.6948
1.6965
1.6993
0.7211
0.7283
0.7336
0.7344
0.7288
0.9113
0.9107
0.9103
0.9106
0.9137
The first value represents the number of detail blocks, and the second value denotes the number of smooth blocks. The average values of eight quality metrics for all test images are shown. The best scores are shown in boldface, and second-best values are marked in italic.
Table 3.
Fusion Performance of the Proposed Method with Different Weight Values of Base Imagesa
Weight Value of Base Images
Quality Metrics
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
6.9813
6.9142
6.8491
6.7997
6.7770
6.7817
6.8106
6.8619
6.9323
13.9625
13.8283
13.6981
13.5993
13.5540
13.5634
13.6212
13.7238
13.8645
0.4295
0.4333
0.4371
0.4398
0.4410
0.4406
0.4388
0.4356
0.4308
0.9002
0.9002
0.9002
0.9000
0.8997
0.8994
0.8989
0.8983
0.8976
0.3928
0.3947
0.3958
0.3961
0.3961
0.3957
0.3950
0.3941
0.3926
0.4351
0.4359
0.4362
0.4362
0.4359
0.4355
0.4349
0.4341
0.4331
0.4171
0.4193
0.4208
0.4221
0.4233
0.4245
0.4257
0.4270
0.4279
1.6468
1.6957
1.7117
1.7104
1.6965
1.6692
1.6217
1.5384
1.3853
0.6902
0.7075
0.7213
0.7302
0.7344
0.7337
0.7290
0.7208
0.7091
0.6455
0.6493
0.6516
0.6525
0.6520
0.6501
0.6468
0.6420
0.6356
0.9031
0.9075
0.9102
0.9113
0.9106
0.9083
0.9043
0.8986
0.8915
Average values of 11 quality metrics for all test images. The best scores are shown in boldface, second-best values are marked in italic, and third-best values are underlined.