Chinese Journal of Liquid Crystals and Displays, Volume. 39, Issue 7, 950(2024)
Single image super-resolution reconstruction based on split-attention networks
Fig. 5. Comparison of ResNet and proposed residual structure.（a）ResNet structure；（b）Proposed residual structure.
Fig. 9. Comparison chart of reconstruction effect of “bird” in Set5
Fig. 10. Comparison chart of reconstruction effect of “ppt3” in Set14
Table 1. Influence of the values of Radix and Cardinality on the performance of the model
View tableView in ArticleTable 1. Influence of the values of Radix and Cardinality on the performance of the model
取值 PSNR SSIM 1s16x 25.988 6 0.723 2 2s16x 25.981 1 0.722 2 1s32x 26.106 5 0.724 0 2s32x 26.009 1 0.721 6 1s64x 26.152 5 0.725 0 2s64x 26.102 2 0.724 3 1s128x 26.152 7 0.727 0 2s128x 26.151 2 0.724 5
Table 2. PSNR value and SSIM value of different module combinations on Set5 data set
View tableView in ArticleTable 2. PSNR value and SSIM value of different module combinations on Set5 data set
Methods PSNR/dB SSIM Baseline 27.896 5 0.806 3 Baseline+ResNeSt 28.372 6 0.816 9 Baseline+SN 28.624 3 0.824 1 Baseline+Focal Frequency Loss 28.125 0 0.810 8 Baseline+Charbonnier 28.555 3 0.825 2 Ours 29.145 0 0.843 0
Table 3. Average PSNR of different SR algorithms on four test sets at 4× magnification factor
View tableView in ArticleTable 3. Average PSNR of different SR algorithms on four test sets at 4× magnification factor
DataSet Scale Bicubic ESPCN SRGAN ESRGAN FASRGAN XLSR Proposed Set5 4 23.173 27.677 27.897 28.543 28.302 29.001 29.145 Set14 4 21.996 25.184 25.389 24.505 24.587 25.954 26.152 BSD100 4 22.710 24.947 25.230 23.946 24.026 25.726 25.748 Urban100 4 19.777 22.570 22.853 22.800 23.009 23.309 23.482
Table 4. Average SSIM of different SR algorithms on four test sets at 4× magnification factor
View tableView in ArticleTable 4. Average SSIM of different SR algorithms on four test sets at 4× magnification factor
DataSet Scale Bicubic ESPCN SRGAN ESRGAN FASRGAN XLSR Proposed Set5 4 0.688 7 0.803 5 0.806 3 0.814 5 0.807 2 0.839 4 0.843 0 Set14 4 0.585 2 0.700 9 0.694 1 0.654 5 0.662 2 0.733 7 0.727 0 BSD100 4 0.571 0 0.675 9 0.670 7 0.621 7 0.624 0 0.693 8 0.695 8 Urban100 4 0.554 1 0.676 5 0.683 2 0.703 8 0.711 3 0.712 6 0.719 9
Get Citation
Copy Citation Text
Yanfei PENG, Lanxi LIU, Gang WANG, Xin MENG, Yongxin LI. Single image super-resolution reconstruction based on split-attention networks[J]. Chinese Journal of Liquid Crystals and Displays, 2024, 39(7): 950
Category: Research Articles
Received: Jun. 28, 2023
Accepted: --
Published Online: Jul. 23, 2024
The Author Email: Lanxi LIU (932134582@qq.com)
