Multi-Scale Dilated Convolutional Neural Network Based Multi-Focus Image Fusion Algorithm

Fig. 7. Fused results of “children” image. (a)(b) Source images; (c) NSCT; (d) SR; (e) IMF; (f) MWGF; (g) CNN; (h) DeepFuse; (i) DenseFuse-ADD; (j) DenseFuse-L1; (k) IFCNN-MAX; (l) MDF-Net

Download full size

Fig. 8. Fusion results of “monkey” image. (a)(b) Source images; (c) NSCT; (d) SR; (e) IMF; (f) MWGF; (g) CNN; (h) DeepFuse; (i) DenseFuse-ADD; (j) DenseFuse-L1; (k) IFCNN-MAX; (l) MDF-Net

Download full size

Fig. 9. Fusion results of “gymnasium” image. (a)(b) Source images; (c) NSCT; (d) SR; (e) IMF; (f) MWGF; (g) CNN; (h) DeepFuse; (i) DenseFuse-ADD; (j) DenseFuse-L1; (k) IFCNN-MAX; (l) MDF-Net

Download full size

Fig. 10. Fusion results of “statue” image. (a)(b) Source images; (c) NSCT; (d) SR; (e) IMF; (f) MWGF; (g) CNN; (h) DeepFuse; (i) DenseFuse-ADD; (j) DenseFuse-L1; (k) IFCNN-MAX; (l) MDF-Net

Download full size

Table 1. Indexes values of various fusion algorithms on "gymnasium" image

View table

Table 1. Indexes values of various fusion algorithms on "gymnasium" image

Algorithm	AG	SF	VIF	$Q^{A B / F}$
NSCT	6.2530	18.7075	0.9100	0.6685
SR	6.2463	18.8690	0.9156	0.7096
IMF	6.2806	18.7851	0.8795	0.6973
MWGF	6.0492	17.4456	0.8422	0.6906
CNN	6.1795	18.5287	0.9268	0.7165
DeepFuse	4.1720	10.9647	0.7062	0.5375
DenseFuse-ADD	4.5071	12.0923	0.8049	0.6110
DenseFuse-L1	4.4546	11.8492	0.7844	0.6084
IFCNN-MAX	6.4207	18.9627	0.9421	0.6876
MDF-Net	6.4541	19.3478	0.9424	0.7098

Table 2. Average indexes values of various fusion algorithms on Lytro dataset

View table

Table 2. Average indexes values of various fusion algorithms on Lytro dataset

Algorithm	AG	SF	VIF	$Q^{A B / F}$
NSCT	6.8506	18.8012	0.9187	0.7168
SR	6.8328	18.9940	0.9309	0.7437
IMF	6.9688	19.2298	0.9319	0.7406
MWGF	6.8446	19.0156	0.9308	0.7437
CNN	6.8814	19.0005	0.9373	0.7555
DeepFuse	4.0751	10.4187	0.6613	0.4860
DenseFuse-ADD	4.4676	11.5175	0.7727	0.5806
DenseFuse-L1	4.3616	11.0373	0.7510	0.5587
IFCNN-MAX	6.8742	18.9718	0.9327	0.7105
MDF-Net	6.9957	19.2976	0.9437	0.7488

Table 3. Indexes results of different variants of MDF-Net
View table
Table 3. Indexes results of different variants of MDF-Net
Variants of MDF-Net AG SF VIF $Q^{A B / F}$
Without-Dilated 6.8088 18.8046 0.9132 0.7485
Without-SE 6.9459 19.1778 0.9372 0.7472
Without-Cat 6.8248 18.8041 0.9213 0.7468
MDF-Net 6.9957 19.2976 0.9437 0.7488

Table 4. Ablation experiment on number of MDB modules
View table
Table 4. Ablation experiment on number of MDB modules
#MDB AG SF VIF $Q^{A B / F}$
1 6.7833 18.7125 0.9217 0.7465
2 6.9219 19.0769 0.9334 0.7473
3 6.9957 19.2976 0.9437 0.7488
4 6.7643 18.6264 0.9125 0.7443

Table 5. Ablation experiment on number of dilated convolution branches in MDB
View table
Table 5. Ablation experiment on number of dilated convolution branches in MDB
#Branch AG SF VIF $Q^{A B / F}$
2 6.7490 18.5700 0.9035 0.7484
3 6.8243 18.7955 0.9169 0.7446
4 6.9957 19.2976 0.9437 0.7488
5 6.9094 18.9495 0.9301 0.7403

Table 6. Ablation experiment on number of feature channels in MDB
View table
Table 6. Ablation experiment on number of feature channels in MDB
#Channel AG SF VIF $Q^{A B / F}$
64-64-64 6.8616 18.8961 0.9269 0.7457
128-128-128 6.8891 18.9124 0.9235 0.7455
256-256-256 6.8319 18.8076 0.9294 0.7401
64-128-256 6.9957 19.2976 0.9437 0.7488

Tools

Get Citation

Copy Citation Text

Haitao Yin, Wei Zhou. Multi-Scale Dilated Convolutional Neural Network Based Multi-Focus Image Fusion Algorithm[J]. Laser & Optoelectronics Progress, 2023, 60(2): 0210003

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites