Author Affiliations
School of Physics and Optoelectronic Engineering, Xiangtan University, Xiangtan, Hunan 411105, Chinashow less
Fig. 1. Schematic diagram of off⁃axis Fresnel digital hologram recording
Fig. 2. Method for off⁃axis Fresnel digital hologram reconstruction based on deep learning
Fig. 3. The structure model of ResNet
Fig. 4. Structure of each module in ResNet model
Fig. 5. Training samples of numerical simulation experiment
Fig. 6. The training loss curves in ResNet training by using L2 loss functions
Fig. 7. Schematic diagram of off⁃axis Fresnel digital hologram recording
Fig. 8. Off⁃axis Fresnel digital hologram nonlinear reconstruction with ResNet
Fig. 9. ResNet reconstruction results of test dateset with different diffraction distances (plane wave training, z0=0.3 m)
Fig. 10. Distance Robust ResNet reconstruction results for different diffraction distance (plane wave)
Fig. 11. The MAE, RMSE, and SSIM of Distance Robust ResNet reconstruction results and corresponding object images for different diffraction distance (plane wave)
| Frequency filtering | Four⁃step phase shift | Plane z0=0.3 m | Plane z0=0.4 m | Plane z0=0.5 m | Spherical z0=0.3 m | Spherical z0=0.4 m | Spherical z0=0.5 m | MAE | 1.870 5 | 1.808 3 | ↓1.212 0 | 1.398 0 | 1.524 7 | 1.932 5 | 2.111 2 | 2.200 9 | RMSE | 2.677 2 | 2.515 0 | ↓2.208 7 | 2.295 0 | 2.427 3 | 2.609 0 | 2.688 5 | 2.675 2 | SSIM | 0.729 1 | 0.740 8 | ↓0.918 0 | 0.909 1 | 0.906 6 | 0.888 6 | 0.883 1 | 0.879 6 |
|
Table 1. Compare MAE, RMSE and SSIM for deep learning object reconstruction with traditional algorithms
| Plane z0=0.3 m | Plane z0=0.4 m | Plane z0=0.5 m | Robust z0=0.3 m | Robust z0=0.4 m | Robust z0=0.5 m | MAE | ↓1.212 0 | 1.398 0 | 1.524 7 | 2.870 8 | 1.744 9 | 1.552 4 | RMSE | ↓2.208 7 | 2.295 0 | 2.427 3 | 3.037 1 | 2.599 3 | 2.463 5 | SSIM | ↑0.918 0 | 0.909 1 | 0.906 6 | 0.889 0 | 0. 891 9 | 0.882 9 | | |
|
Table 2. Compare MAE, RMSE and SSIM for distance⁃robust with single diffraction distance ResNet reconstruction results