notesum.ai
Published at November 15Rethinking Normalization Strategies and Convolutional Kernels for Multimodal Image Fusion
cs.CV
cs.AI
Released Date: November 15, 2024
Authors: Dan He1, Guofen Wang2, Weisheng Li1, Yucheng Shu1, Wenbo Li1, Lijian Yang1, Yuping Huang1, Feiyan Li1
Aff.: 1Chongqing University of Posts and Telecommunications; 2Chongqing Normal University

| Dataset: MRI-CT Image Fusion (50 pairs) [42] | ||||||
| Methods | SD | AG | SF | SCD | VIFF | SSIM |
| CDDFu* [61] | 84.803 | 9.292 | 38.121 | 1.376 | 0.413 | 0.727 |
| EMMA* [63] | 86.493 | 7.534 | 27.405 | 1.464 | 0.421 | 0.668 |
| MMDRF* [5] | 75.768 | 5.723 | 22.993 | 1.263 | 0.371 | 0.263 |
| GeseNet [20] | 85.591 | 9.360 | 37.552 | 1.358 | 0.405 | 0.529 |
| FATFus [39] | 72.680 | 7.688 | 26.901 | 0.906 | 0.119 | 0.717 |
| Ours | 94.033 | 8.242 | 33.836 | 1.576 | 0.456 | 0.731 |
| Dataset: MRI-PET Image Fusion (50 pairs) [42] | ||||||
| Methods | SD | AG | SF | SCD | VIFF | SSIM |
| CDDFu* [61] | 69.309 | 8.257 | 27.883 | 1.405 | 0.502 | 0.739 |
| EMMA* [63] | 74.011 | 7.151 | 22.547 | 1.602 | 0.555 | 0.668 |
| MMDRF* [5] | 71.833 | 6.131 | 20.099 | 1.604 | 0.539 | 0.262 |
| GeseNet [20] | 71.110 | 8.311 | 27.508 | 1.470 | 0.553 | 0.486 |
| FATFus [39] | 69.284 | 8.760 | 30.089 | 1.278 | 0.402 | 0.727 |
| Ours | 78.345 | 9.424 | 31.663 | 1.660 | 0.590 | 0.734 |
| Dataset: MRI-SPECT Image Fusion (50 pairs) [42] | ||||||
| Methods | SD | AG | SF | SCD | VIFF | SSIM |
| CDDFu* [61] | 66.015 | 7.364 | 24.824 | 1.039 | 0.521 | 0.719 |
| EMMA* [63] | 68.472 | 6.378 | 20.555 | 1.528 | 0.606 | 0.664 |
| MMDRF* [5] | 68.655 | 5.499 | 18.248 | 1.597 | 0.589 | 0.263 |
| GeseNet [20] | 66.965 | 7.469 | 24.686 | 1.409 | 0.602 | 0.490 |
| FATFus [39] | 73.948 | 7.988 | 27.301 | 1.431 | 0.561 | 0.711 |
| Ours | 75.481 | 8.731 | 29.770 | 1.651 | 0.637 | 0.723 |