Acta Photonica Sinica, Volume. 51, Issue 4, 0410001(2022)
Polarization Image Super-resolution Reconstruction Based on Dual Attention Residual Network
Figures & Tables(18)
Fig. 4. Network architecture of dual attention deep residual network
Fig. 10. Reconstruction effect of truck contraction model by different methods
Fig. 11. Reconstruction effect of airplane image by different methods
Fig. 14. Comparison between polarization image reconstruction results of camouflage net and the system-acquired results
Fig. 15. Curve of relationship between no reference evaluation index and supercell multiple
Table 1. Comparison of indicators of fabric image polarization by different algorithms
View tableView in ArticleTable 1. Comparison of indicators of fabric image polarization by different algorithms
Scale Method Params 0° 60° 120° PSNR/dB SSIM PSNR/dB SSIM PSNR/dB SSIM 2 Bicubic - 37.626 0 0.939 3 34.775 4 0.925 4 36.116 4 0.929 1 SRCNN 57K 39.808 2 0.959 0 37.164 2 0.950 7 38.393 5 0.952 4 FSRCNN 12K 39.858 2 0.959 9 37.520 4 0.952 4 38.583 1 0.953 8 EDSR 40.7M 42.277 4 0.968 6 39.892 0 0.962 2 40.974 9 0.963 6 Ours 17.7M 42.294 5 0.968 8 39.919 2 0.962 6 40.988 0 0.963 9 3 Bicubic - 34.056 6 0.876 1 31.197 2 0.846 1 32.629 7 0.857 4 SRCNN 57K 35.727 5 0.902 9 32.984 9 0.881 0 34.356 1 0.888 8 FSRCNN 12K 35.572 2 0.902 7 33.047 5 0.881 8 34.280 4 0.889 0 EDSR 43.6M 37.783 2 0.923 6 35.145 5 0.905 0 36.403 2 0.911 5 Ours 17.9M 37.834 8 0.924 6 35.245 3 0.906 3 36.494 2 0.912 8 4 Bicubic - 31.966 8 0.824 3 29.106 6 0.780 0 30.555 3 0.798 3 SRCNN 57K 33.382 4 0.848 7 30.478 2 0.811 4 31.910 6 0.826 3 FSRCNN 12K 33.471 3 0.849 3 30.665 4 0.813 7 32.031 8 0.827 7 EDSR 43M 35.272 3 0.881 6 32.495 8 0.850 5 33.794 3 0.862 4 Ours 17.9M 35.356 8 0.882 5 32.563 5 0.851 9 33.863 4 0.863 5
Table 2. Comparison of indicators on the analysis of monument image full polarization by different algorithms
View tableView in ArticleTable 2. Comparison of indicators on the analysis of monument image full polarization by different algorithms
Method Time/s Image V Image Ey Image S0 Image Q Image U PSNR/dB SSIM PSNR/dB SSIM PSNR/dB SSIM PSNR/dB SSIM PSNR/dB SSIM Bicubic 0.05 22.406 9 0.907 0 22.043 2 0.909 2 52.669 6 0.997 5 20.953 3 0.783 8 22.587 6 0.909 6 SRCNN 0.24 27.944 3 0.963 6 26.216 7 0.965 9 51.989 4 0.997 6 23.089 1 0.870 9 28.221 5 0.964 3 FSRCNN 0.25 27.567 7 0.960 4 25.850 3 0.958 8 51.167 9 0.997 3 24.194 1 0.903 8 27.786 7 0.961 9 EDSR 0.49 32.761 5 0.990 8 29.315 7 0.982 1 53.533 4 0.997 6 29.268 5 0.958 5 32.860 4 0.990 9 Ours 0.48 35.331 9 0.993 9 30.940 4 0.987 1 53.538 1 0.997 6 29.141 3 0.956 1 35.052 7 0.993 7
Table 3. No reference comparison of camouflage net polarization images results
View tableView in ArticleTable 3. No reference comparison of camouflage net polarization images results
Scale Information entropy Definition LR HR SR LR HR SR 6.83 6.87 6.85 0.91 0.93 0.93 6.57 6.61 6.59 0.91 0.93 0.94 6.72 6.76 6.74 0.91 0.93 0.94
Tools
Get Citation
Copy Citation Text
Guoming XU, Jie WANG, Jian MA, Yong WANG, Jiaqing LIU, Yi LI. Polarization Image Super-resolution Reconstruction Based on Dual Attention Residual Network[J]. Acta Photonica Sinica, 2022, 51(4): 0410001
Paper Information
Category:
Received: Sep. 18, 2021
Accepted: Nov. 17, 2021
Published Online: May. 18, 2022
The Author Email: WANG Jie (1197545193@qq.com)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20