Author Affiliations
1National Key Laboratory of Optical Field Manipulation Science and Technology, Chinese Academy of Sciences, Chengdu, Sichuan 610209, China2Key Laboratory of Beam Control, Chinese Academy of Sciences, Chengdu, Sichuan 610209, China3Institute of Optics and Electronics, Chinese Academy of Sciences, Chengdu, Sichuan 610209, China4University of Chinese Academy of Sciences, Beijing 100049, Chinashow less
Fig. 1. The image degradation process
Fig. 2. Schematic diagram of up-sampling based on micro-scanning
Fig. 3. Three ways of micro-scanning
Fig. 4. Schematic diagram of reconstruction based on micro-scanning imaging
Fig. 5. Flow chart of our algorithm
Fig. 6. Schematic diagram of the experimental setup
Fig. 7. Schematic diagram of selection module. Left: Image sequence; Right: An image grid with complete sub-pixel information
Fig. 8. Four cases of displacement. (a) Four possible cases of pixel shift; (b) Four modes of integer pixel shift
Fig. 9. Schematic diagrams of information extraction in four integer pixel shift cases
Fig. 10. Schematic diagram of denoise module. (a) Schematic of matching same pixel of multiple images; (b) Pixel value and noise points (red circle) of same pixels
Fig. 11. Experiment sets of the active displacement imaging method
Fig. 12. Camera position (red point)
Fig. 13. Comparison result between ground truth and calculation. (a) Comparison result at 25 points; (b) Comparison of error at 25 points
Fig. 14. Super-resolution reconstruct results of different algorithms at scale of 4. (a) MFPOCS
[20]; (b) ACNet
[6]; (c) Ours
Fig. 15. MTF curves of different algorithms at different scales
Fig. 16. Original pictures and their ROI (red rectangle). (a) Simple image; (b) Complex image; (c) Panda image
Fig. 17. Comparison of the traditional interpolation and our interpolation at 4 times. (a) Ground truth; (b) Ours; (c) Linear; (d) Bicubic
Fig. 18. Super-resolution reconstruction results of simple image at different scales using the algorithms of MFPOCS
[20](yellow rectangle), ACNet
[6] (green rectangle) and ours (red rectangle)
Fig. 19. Super-resolution reconstruction results of ROI of simple image at different scales using the algorithm of MFPOCS
[20] (yellow rectangle), ACNet
[6] (green rectangle) and ours (red rectangle)
Fig. 20. Super-resolution results of the complex image using the algorithms of MFPOCS
[20] (yellow rectangle), ACNet
[6] (green rectangle) and ours (red rectangle)
Fig. 21. Super-resolution reconstruction results of panda image at different scales using the algorithms of MFPOCS
[20](yellow rectangle), ACNet
[6](green rectangle) and ours (red rectangle)
Fig. 22. Super-resolution reconstruction results of the complex image at different scales using the algorithms of MFPOCS
[20] (yellow rectangle), ACNet
[6] (green rectangle) and ours (red rectangle)
Fig. 23. Super-resolution reconstruction results of ROI of panda image at different scales using the algorithms of MFPOCS
[20](yellow rectangle), ACNet
[6] (green rectangle) and ours (red rectangle)
Images | Scale | 2 | 3 | 4 | 5 | Simple image | POCS[20] | 0.8493 | 0.8093 | 0.8568 | 0.8330 | ACNet[6] | 0.9876 | 0.9764 | 0.9623 | 0.9418 | Ours | 0.9921 | 0.9694 | 0.9949 | 0.9778 | Complex image | MFPOCS[20] | 0.6838 | 0.6539 | 0.6880 | 0.6714 | ACNet[6] | 0.9462 | 0.8926 | 0.8051 | 0.7358 | Ours | 0.9517 | 0.9183 | 0.9592 | 0.9250 | Panda | MFPOCS[20] | 0.6263 | 0.6187 | 0.5748 | 0.5653 | ACNet[6] | 0.7046 | 0.6789 | 0.6014 | 0.5736 | Ours | 0.6696 | 0.6215 | 0.6255 | 0.6002 |
|
Table 1. SSIM of three algorithms
Images | Scale | 2 | 3 | 4 | 5 | Simple image | MFPOCS[20] | 29.4390 | 26.4232 | 29.4868 | 26.4090 | ACNet[6] | 47.7125 | 43.6358 | 39.2593 | 36.5734 | Ours | 46.7883 | 45.5723 | 43.9699 | 39.1457 | Complex Image | MFPOCS[20] | 20.2672 | 20.1293 | 20.1270 | 20.1398 | ACNet[6] | 29.1521 | 27.7453 | 24.1961 | 21.9834 | Ours | 27.4424 | 29.5396 | 28.9332 | 26.6562 | Panda | MFPOCS[20] | 24.0725 | 22.3215 | 20.4376 | 19.8857 | ACNet[6] | 25.7617 | 23.5169 | 19.5048 | 18.3985 | Ours | 24.0031 | 23.1915 | 21.9751 | 21.7718 |
|
Table 2. PSNR of three algorithms
Images | Scale | 2 | 3 | 4 | 5 | Simple image | MFPOCS[20] | 314.7994 | 211.4553 | 131.7388 | 105.3359 | ACNet [6] | 338.4507 | 294.9276 | 201.8644 | 145.9228 | Ours | 320.5050 | 265.3140 | 215.9100 | 184.1190 | Complex Image | MFPOCS[20] | 350.7845 | 242.5359 | 162.4356 | 129.1925 | ACNet [6] | 471.1172 | 395.2651 | 275.1865 | 216.5397 | Ours | 446.9067 | 383.5727 | 350.1874 | 314.0308 | Panda | MFPOCS[20] | 214.7590 | 147.4026 | 91.8175 | 77.1331 | ACNet [5] | 271.9497 | 263.0891 | 191.2695 | 163.3797 | Ours | 253.5610 | 389.1927 | 497.7272 | 205.6040 |
|
Table 3. Mean gradient of three algorithms