Chinese Journal of Liquid Crystals and Displays, Volume. 37, Issue 11, 1476(2022)
Fourier ptychography based on multi-scale feature fusion network
Fig. 7. Convergence of DCCA module and standard convolution in training
Fig. 8. Reconstruction on simulation datasets of each method with different overlap ratio.（a）25%；（b）50%.
Fig. 9. Robustness analysis of different methods to Gaussian noise.（a）PSNR；（b）SSIM.
Table 1. Average PSNR and SSIM on the test datasets for networks with 5 feature map numbers
View tableView in ArticleTable 1. Average PSNR and SSIM on the test datasets for networks with 5 feature map numbers
Number of feature map Max feature map size Min feature map size PSNR/dB SSIM 2 128×128 64×64 23.70 0.781 3 3 128×128 32×32 25.13 0.816 9 4 128×128 16×16 28.87 0.847 1 5 128×128 8×8 27.39 0.832 2 6 128×128 4×4 27.21 0.817 3
Table 2. PSNR and SSIM of different module on test datasets
View tableView in ArticleTable 2. PSNR and SSIM of different module on test datasets
Module Set5 Set14 B100 Urban100 PSNR/dB SSIM PSNR/dB SSIM PSNR/dB SSIM PSNR/dB SSIM Conv 32.84 0.938 1 28.93 0.863 8 28.18 0.839 5 28.38 0.840 4 ConvCA 33.17 0.940 3 29.12 0.864 9 28.35 0.833 4 28.81 0.842 9 DCCA 33.46 0.941 5 30.05 0.867 3 28.42 0.839 4 28.93 0.847 1
Table 3. PSNR and SSIM of each method with different overlap ratio
View tableView in ArticleTable 3. PSNR and SSIM of each method with different overlap ratio
method Overlap ratio/% Set5 Set14 B100 Urban100 PSNR/dB SSIM PSNR/dB SSIM PSNR/dB SSIM PSNR/dB SSIM PtychNet 25 19.42 0.612 3 16.99 0.546 2 16.21 0.474 9 15.90 0.431 6 AP 25 25.95 0.820 2 23.43 0.771 4 21.96 0.728 6 21.20 0.701 1 FPNN 25 23.62 0.735 8 22.71 0.699 7 22.72 0.714 7 20.79 0.675 0 FPMFN 25 29.07 0.891 7 27.91 0.833 1 27.48 0.835 7 26.55 0.828 4 PtychNet 50 25.49 0.805 1 24.29 0.764 9 23.35 0.638 4 22.81 0.644 1 AP 50 30.92 0.915 8 28.01 0.841 3 27.39 0.829 5 26.53 0.824 9 FPNN 50 28.92 0.914 3 28.78 0.849 7 24.49 0.801 3 24.75 0.796 2 FPMFN 50 33.46 0.941 5 29.75 0.867 3 28.42 0.839 4 28.93 0.847 1 PtychNet 62.5 27.03 0.836 7 26.31 0.818 7 23.18 0.748 7 22.14 0.721 3 AP 62.5 32.25 0.952 3 29.40 0.880 7 28.57 0.862 1 27.72 0.832 3 FPNN 62.5 31.40 0.929 6 28.26 0.855 2 26.64 0.820 3 26.24 0.831 4 FPMFN 62.5 35.18 0.968 4 29.39 0.894 7 28.71 0.864 2 29.38 0.873 9 PtychNet 70 27.88 0.855 8 26.26 0.836 8 25.04 0.782 4 24.58 0.747 5 AP 70 33.92 0.958 8 31.94 0.907 4 28.36 0.887 4 30.41 0.905 0 FPNN 70 33.69 0.941 2 29.34 0.880 2 27.69 0.848 3 29.68 0.883 8 FPMFN 70% 36.23 0.963 3 32.06 0.931 9 29.35 0.904 7 31.07 0.912 8
Table 4. Time consumption of reconstructing 128×128 image in different methods
View tableView in ArticleTable 4. Time consumption of reconstructing 128×128 image in different methods
Overlap ratio/% PtychNet AP FPNN FPMFN 25 17.3 532.2 2 659.7 19.4 50 32.4 1 181.5 8 419.9 35.1 62.5 52.7 2 057.1 17 374.2 58.3 70 83.2 2 506.4 28 587.4 90.5
Get Citation
Copy Citation Text
Dong-han SONG, Bin WANG, You-qiang ZHU, Xin LIU. Fourier ptychography based on multi-scale feature fusion network[J]. Chinese Journal of Liquid Crystals and Displays, 2022, 37(11): 1476
Category: Research Articles
Received: Mar. 23, 2022
Accepted: --
Published Online: Nov. 3, 2022
The Author Email: Bin WANG (175969722@qq.com)
