Author Affiliations
1College of Information and Control Engineering, Xi'an University of Architecture and Technology, Xi'an 710055, Shaanxi, China2Shaanxi Institute for the Preservation of Cultural Heritage, Xi'an 710075, Shaanxi, Chinashow less
Fig. 1. X-ray imaging effect diagrams of copper mirror at different shooting energy. (a) High-energy shooting; (b) low-energy shooting
Fig. 2. Copper mirror generative confrontation fusion network framework
Fig. 3. Generator network structure based on multi-scale feature fusion
Fig. 4. Discriminator network structure
Fig. 5. Network training process
Fig. 6. Copper mirror X-ray image control group. (a) High energy shooting effect group; (b) low energy shooting effect group
Fig. 7. Comparison of copper mirror fusion by different algorithms. (a) LP; (b) GFF; (c) LE-LP; (d) BFWLS; (e) O-BFWLS
Fig. 8. Comparison of copper mirror fusion in ablation experiment. (a) Fusion-GAN-CLF; (b) Fusion-GAN-MSFF; (c) proposed algorithm
Fig. 9. Comparison of fusion effect details. (a) High-energy X-source image; (b) low-energy X-source image; (c) Fusion-GAN-CLF; (d) Fusion-GAN-MSF; (e) proposed algorithm
| Image | Metric | LP | GFF | LE-LP | BFWLS | O-BFWLS |
|---|
| Group 1 | EN SF AG CEN JE NFMI | 4.9667 10.6807 2.0742 2.566/2.971 7.915/7 1.9646 | 4.857 10.8896 2.2773 0.590/6.059 7.584/7.092 2.1598 | 4.8333 9.9375 1.9118 1.227/5.112 7.590/7 2.0651 | 4.4198 9.1235 1.5558 0.244/6.315 6.597/6.447 2.5458 | 4.4291 9.1596 1.6132 0.204/6.341 6.576/6.455 2.5696 | | Group 2 | EN SF AG CEN JE NFMI | 4.2420 6.8456 2.0520 0.193/0.303 7.175/5.809 1.9256 | 4.1233 7.4411 1.9256 0.525/0.457 6.882/5.936 2.0003 | 4.205 7.2289 1.9665 0.467/0.469 6.879/5.918 2.1584 | 4.4404 6.2832 1.5032 0.538/0.562 6.100/5.763 2.856 | 4.2231 6.2832 1.5032 0.531/0.541 5.773/5.759 3.138 | | Group 3 | EN SF AG CEN JE NFMI | 5.1207 6.8398 1.8539 0.047/0.550 7.3246/7 3.0145 | 5 5.8605 1.8432 0.455/0.590 8.3765/7 2.1592 | 4.9781 5.8595 1.7057 1.230/0.448 7.8212/7 2.7046 | 5.1144 6.5366 1.4828 0.052/0.564 7.2544/7 3.2308 | 5.1144 6.5366 1.4828 1.695/0.571 5.0175/7 4.9991 | | Group 4 | EN SF AG CEN JE NFMI | 6.0487 6.4464 1.5824 1.411/1.095 11.36/9.669 1.1782 | 5.6106 8.6505 1.7474 3.461/0.806 10.86/9.155 0.9797 | 5.5615 8.2303 1.6409 1.364/1.266 10.70/8.996 0.8909 | 5.4072 8.3238 1.3300 0.097/1.631 10.70/8.995 1.002 | 5.4152 8.1558 1.4989 0.065/1.483 10.73/9.021 0.9821 | | Group 5 | EN SF AG CEN JE NFMI | 3.9007 5.5334 1.3779 0.437/0.338 6.177/5.645 1.8198 | 3.6696 6.0313 1.2521 0.202/0.193 5.901/5.443 1.8778 | 3.7686 5.5984 1.2257 0.700/0.340 5.658/5.343 2.1215 | 3.735 6.0035 1.1408 0.149/0.544 5.126/5.198 2.4949 | 3.7331 6.0789 1.2949 0.058/0.541 4.623/5.197 2.9579 |
|
Table 1. Objective evaluation results of different algorithms on different images
| Image | Metric | Fusion-GAN-CLF | Fusion-GAN-MSFF | Proposed algorithm |
|---|
| Group 1 | EN SF AG CEN JE NFMI | 4.5475 10.399 2.1427 9.493/1.006 7.832/6.872 2.9488 | 4.5868 10.0421 1.9867 9.455/1.278 7.770/6.839 3.0166 | 4.5875 10.7134 2.222 9.657/1.021 7.858/6.905 2.9875 | | Group 2 | EN SF AG CEN JE NFMI | 4.2864 8.6157 2.1602 12.37/15.60 6.865/5.250 2.3429 | 4.2724 8.4269 2.0118 12.43/15.71 6.839/5.252 2.3629 | 4.3101 8.7189 2.1705 12.33/15.55 6.914/5.259 2.3559 | | Group 3 | EN SF AG CEN JE NFMI | 5.942 10.2126 3.4667 9.674/13.42 8.055/6.836 2.537 | 5.933 9.4506 3.0807 9.447/13.35 7.898/6.800 2.5786 | 5.9653 10.3992 3.5488 9.438/13.23 8.055/6.846 2.6648 | | Group 4 | EN SF AG CEN JE NFMI | 6.3531 7.5210 1.7487 5.848/12.71 10.60/8.908 0.9464 | 6.3967 7.7080 1.6680 6.021/12.72 10.65/8.958 0.9635 | 6.4849 7.7082 1.7529 6.246/12.75 10.75/9.052 0.9989 | | Group 5 | EN SF AG CEN JE NFMI | 3.7859 8.6920 1.6906 12.75/15.63 5.768/4.910 2.2131 | 3.7733 8.5561 1.6821 12.71/15.65 5.748/4.893 2.2188 | 3.7878 9.3739 1.6934 12.74/15.63 6.262/5.685 2.2303 |
|
Table 2. Comparison of experimental data of different algorithms
| Level | Hinder the scale | Score |
|---|
| Excellent | Observe all diseases and identify types | 5 | | Good | Observe all diseases and identify some types | 4 | | Moderate | Observe part of the disease and identify the type | 3 | | Pass | Observe part of the disease but cannot distinguish the type | 2 | | Poor | Unable to observe the disease | 1 |
|
Table 3. Subjective score of experts in cultural relic restoration
| Group | Fusion-GAN-CLF | Fusion-GAN-MSFF | Proposed algorithm |
|---|
| Group 1 | 4 | 3.8 | 4.7 | | Group 2 | 4.2 | 4.4 | 4.8 | | Group 3 | 3.8 | 3.9 | 4.4 | | Group 4 | 3.6 | 3.4 | 4.2 | | Group 5 | 4.2 | 3.9 | 4.6 |
|
Table 4. Subjective evaluation results
