Acta Optica Sinica, Volume. 39, Issue 2, 0210003(2019)
Super-Resolution Reconstruction of Accelerated Image Based on Deep Residual Network
Figures & Tables(15)
Fig. 5. (a) Variation of loss function of 12-layer network with number of iterations; (b) variation of PSNR average value of set 5 with number of iterations under different layers
Fig. 6. Variation of PSNR average value of set 5 under different activation functions with number of iterations
Fig. 7. Relationship between running time and PSNR average value of set 5 under different algorithms
Fig. 8. Variation of PSNR average value of set 5 under different optimization methods with number of iterations
Fig. 9. Variation of PSNR average value of set 5 under different filter numbers with number of iterations
Fig. 10. Variation of PSNR average value of set 5 under different network models with number of iterations. (a) Networks of 6-layer and 8-layer; (b) networks of 10-layer and 12-layer
Table 1. PSNR/SSIM average values of test sets at different depths
View tableTable 1. PSNR/SSIM average values of test sets at different depths
Depth PSNR average value SSIM average value set 5 set 14 set 5 set 14 6 33.48 29.65 0.9249 0.8936 10 33.58 29.67 0.9265 0.8941 12 33.60 29.69 0.9268 0.8944
Table 2. PSNR average value of set 5, set 14, and BSD100 under different algorithms
View tableTable 2. PSNR average value of set 5, set 14, and BSD100 under different algorithms
Data set Scale Bicubic ScSR NE+LLE ANR SRCNN ESPCN DRSR ×2 33.65 35.13 35.76 35.83 36.36 36.39 37.41 set 5 ×3 30.42 31.54 31.91 32.00 32.52 32.78 33.60 ×4 28.44 28.24 29.66 29.74 30.15 30.21 31.18 ×2 30.21 31.36 31.78 31.81 32.21 32.21 32.95 set 14 ×3 27.51 28.36 28.59 28.64 29.03 29.13 29.69 ×4 25.97 25.97 26.78 26.83 27.23 27.17 27.83 ×2 29.41 29.41 30.39 30.43 30.89 30.93 31.55 BSD100 ×3 27.07 27.67 27.78 27.81 28.11 28.27 28.54 ×4 25.84 25.84 26.39 26.41 26.63 26.59 27.00
Table 3. SSIM average value of set 5, set 14, and BSD100 under different algorithms
View tableTable 3. SSIM average value of set 5, set 14, and BSD100 under different algorithms
Data set Scale Bicubic ScSR NE+LLE ANR SRCNN ESPCN DRSR ×2 0.9355 0.9428 0.9537 0.9546 0.9566 0.9568 0.9621 set 5 ×3 0.8779 0.8851 0.9053 0.9062 0.9129 0.9162 0.9268 ×4 0.8185 0.8025 0.8516 0.8533 0.8621 0.8578 0.8846 ×2 0.9348 0.9564 0.9575 0.9578 0.9591 0.9598 0.9630 set 14 ×3 0.8494 0.8649 0.8802 0.8777 0.8846 0.8874 0.8941 ×4 0.7799 0.7800 0.8152 0.8173 0.8206 0.8225 0.8361 ×2 0.8401 0.8702 0.8729 0.8742 0.8827 0.8832 0.8924 BSD100 ×3 0.7301 0.7514 0.7702 0.7718 0.7778 0.7816 0.7917 ×4 0.6470 0.6470 0.6875 0.6896 0.6919 0.6942 0.7092
Tools
Get Citation
Copy Citation Text
Zhihong Xi, Caiyan Hou, Kunpeng Yuan, Zhuoqun Xue. Super-Resolution Reconstruction of Accelerated Image Based on Deep Residual Network[J]. Acta Optica Sinica, 2019, 39(2): 0210003
Paper Information
Category: Image Processing
Received: May. 3, 2018
Accepted: Sep. 25, 2018
Published Online: May. 10, 2019
The Author Email:
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20