Author Affiliations
College of Information Science and Technology, Jinan University, Guangzhou, Guangdong 510632, Chinashow less
Fig. 1. Diagram of a typical FPM system
Fig. 2. Examples of 7 × 7 low-resolution images acquired by FPM. (a) Original image; (b) Fourier spectrogram; (c) low-resolution image; (d) images observed by LEDs in different positions
Fig. 3. Structure diagram of proposed Fourier stacked microscopy imaging network model
Fig. 4. Convolution module
Fig. 5. Upsampling reconstruction module
Fig. 6. Residual dense module with channel attention mechanism
Fig. 7. Channel attention module
Fig. 8. Example graph for building and using training datasets
Fig. 9. Schematic of three LED lighting modes. (a) Turn on all LEDs; (b) turn on bright-field area LEDs; (c) turn on rhombus area LEDs
Fig. 10. Convergence analysis of FPRDCA with different values of N, I, and G and three sampling patterns. (a) N; (b) I; (c) G; (d) three sampling patterns
Fig. 11. Convergence analysis of FPRDCA, FPRD, and PtychNet
Fig. 12. Test results of each method on the dataset
Method | Set5 | Set14 | Urban100 | Manga109 | B100 |
---|
PSNR /dB | SSIM | PSNR /dB | SSIM | PSNR /dB | SSIM | PSNR /dB | SSIM | PSNR /dB | SSIM |
---|
PtychNet | 25.33 | 0.8088 | 24.32 | 0.7209 | 22.46 | 0.6752 | 23.36 | 0.7966 | 23.58 | 0.6605 | AP | 30.19 | 0.9124 | 28.36 | 0.8424 | 26.09 | 0.8253 | 29.64 | 0.9012 | 27.03 | 0.8227 | BF-9 | 29.05 | 0.8703 | 26.75 | 0.7628 | 24.76 | 0.7286 | 27.86 | 0.8676 | 24.87 | 0.6869 | 14-4-64 | 33.17 | 0.9339 | 29.87 | 0.8582 | 28.07 | 0.8463 | 32.22 | 0.9366 | 28.01 | 0.8309 | 8-8-64 | 33.21 | 0.9355 | 29.94 | 0.8524 | 28.06 | 0.8399 | 32.23 | 0.9295 | 28.02 | 0.8250 | 20-8-64 | 33.57 | 0.9361 | 29.97 | 0.8693 | 28.18 | 0.8476 | 32.38 | 0.9380 | 28.16 | 0.8402 | *14-8-64 | 33.18 | 0.9351 | 29.86 | 0.8596 | 27.99 | 0.8379 | 32.16 | 0.9282 | 27.88 | 0.8312 | Rhombus | 33.16 | 0.9354 | 29.83 | 0.8577 | 28.07 | 0.8460 | 32.05 | 0.9369 | 27.87 | 0.8283 | 14-8-16 | 33.02 | 0.9319 | 29.80 | 0.8581 | 27.94 | 0.8349 | 31.94 | 0.9247 | 27.93 | 0.8304 | 14-8-32 | 33.33 | 0.9354 | 29.91 | 0.8615 | 28.10 | 0.8398 | 32.26 | 0.9285 | 28.09 | 0.8306 | 14-8-64 | 33.46 | 0.9354 | 29.97 | 0.8602 | 28.19 | 0.8486 | 32.27 | 0.9384 | 28.11 | 0.8319 |
|
Table 1. Average PSNR and SSIM of each model on 5 benchmark datasets
Method | AP | PtychNet | 14-8-64 | 8-8-64 | 20-8-64 | 14-8-32 | 14-4-64 | 14-8-16 |
---|
Time /ms | 730 | 20.1 | 44.4 | 30.2 | 51.7 | 43.4 | 28.8 | 39.6 |
|
Table 2. Average running time for reconstructing a single 128 × 128 image with different models based on 5 benchmark datasets