Optics and Precision Engineering, Volume. 32, Issue 14, 2311(2024)
Iterative reconstruction of compressive sensing combining image hierarchical-feature
Figures & Tables(13)
Fig. 6. Comparison of visual effects of different algorithms for reconstructing images at low sampling rates
Fig. 7. Reconfiguration performance with different iterative stages and mixed architectural layers at 0.10 sampling rate
Fig. 8. Variation curves of Loss with training epoch at different sampling rates
Fig. 9. Hierarchical-feature map of SSCM in different iterative stages at 0.25 sampling rate
Table 1. Comparison of PSNR（dB）/SSIM of reconstructed images on Set11 dataset with different algorithms for each sampling rate
View tableView in ArticleTable 1. Comparison of PSNR（dB）/SSIM of reconstructed images on Set11 dataset with different algorithms for each sampling rate
算法 采样率 0.01 0.04 0.10 0.25 0.50 平均 CSNet+ 20.16/0.517 4 24.59/0.760 4 28.29/0.873 9 33.16/0.942 6 38.92/0.978 9 29.02/0.814 6 AMPNet 20.20/0.558 1 25.26/0.772 2 29.40/0.877 9 34.63/0.948 1 40.33/0.980 4 29.96/0.827 3 ISTANet++ 17.92/0.455 4 22.88/0.672 3 28.34/0.853 1 32.77/0.927 4 38.73/0.972 7 28.13/0.776 2 OPINENet+ 20.02/0.536 2 25.52/0.787 9 29.81/0.890 4 34.81/0.951 4 40.19/0.980 0 30.07/0.829 2 TransCS 20.23/0.504 9 25.41/0.788 3 29.54/0.887 7 35.06/0.954 8 40.49/0.981 5 30.15/0.823 4 CSformer 21.58/0.607 5 26.28/0.806 1 29.69/0.888 2 34.80/0.952 4 40.72/0.982 4 30.61/0.847 3 CHFNet 21.60/0.608 2 26.40/0.812 4 30.10/0.894 9 35.33/0.956 1 40.62/0.981 8 30.81/0.850 7
Table 2. Comparison of PSNR（dB）/SSIM of reconstructed images on different datasets with different algorithms for each sampling rate
View tableView in ArticleTable 2. Comparison of PSNR（dB）/SSIM of reconstructed images on different datasets with different algorithms for each sampling rate
数据集 算法 采样率 0.01 0.04 0.10 0.25 0.50 平均 BSD68 CSNet+ 21.76/0.509 0 24.89/0.676 8 27.39/0.802 5 31.03/0.908 3 35.87/0.967 6 28.19/0.772 8 AMPNet 22.28/0.538 7 25.26/0.676 0 27.86/0.792 6 31.74/0.904 8 36.82/0.968 0 28.79/0.776 0 ISTANet++ 19.89/0.462 1 23.39/0.601 1 26.39/0.741 6 29.79/0.862 0 34.92/0.951 3 26.88/0.723 6 OPINENet+ 21.88/0.516 2 25.16/0.684 1 27.81/0.804 0 31.50/0.906 2 36.32/0.965 8 28.53/0.775 3 TransCS 21.55/0.493 3 25.28/0.688 1 27.86/0.808 5 31.74/0.912 1 36.81/0.969 9 28.65/0.774 4 CSformer 22.81/0.556 6 25.73/0.695 4 28.05/0.803 5 31.80/0.910 6 37.13/0.971 6 29.10/0.787 5 CHFNet 22.94/0.566 1 25.85/0.702 1 28.36/0.827 5 31.96/0.913 9 36.93/0.970 3 29.21/0.796 0 Urban100 CSNet+ 18.47/0.393 2 21.51/0.630 6 24.46/0.798 0 28.75/0.914 6 34.05/0.970 3 25.45/0.741 3 AMPNet 18.85/0.434 5 21.93/0.641 2 25.28/0.804 9 30.39/0.925 8 36.45/0.976 9 26.58/0.756 7 ISTANet++ 16.79/0.347 2 20.04/0.543 4 24.48/0.773 2 28.83/0.894 9 35.99/0.972 7 25.23/0.706 3 OPINENet+ 18.62/0.418 9 22.07/0.666 5 25.77/0.829 7 30.96/0.935 1 36.93/0.979 7 26.87/0.766 0 TransCS 18.32/0.379 5 22.20/0.675 6 25.87/0.834 7 31.14/0.938 7 37.23/0.981 4 26.95/0.762 0 CSformer 19.43/0.465 3 22.62/0.682 3 25.80/0.826 7 30.49/0.932 9 36.45/0.979 3 26.96/0.777 3 CHFNet 19.50/0.466 9 22.46/0.681 7 26.70/0.850 2 31.80/0.941 8 37.48/0.984 1 27.59/0.784 9
Table 3. Comparison of the complexity of different algorithms
View tableView in ArticleTable 3. Comparison of the complexity of different algorithms
采样率 算法 网络
参数量
平均重构时间（GPU）/s Set11 BSD68 Urban100 0.10 TransCS 1.444 M 0.706 2 0.669 8 0.726 7 CSformer 6.647 M 2.772 0 3.582 3 4.804 1 CHFNet 0.608 M 0.027 3 0.031 3 0.029 1 0.50 TransCS 2.284 M 0.380 0 0.705 8 0.711 9 CSformer 6.713 M 2.545 6 3.773 3 4.632 6 CHFNet 1.028 M 0.027 8 0.027 8 0.030 5
Table 4. Ablation experiments with different strategies at 0.50 sampling rate
View tableView in ArticleTable 4. Ablation experiments with different strategies at 0.50 sampling rate
重构策略 测试集的平均PSNR/dB MFEM SSCM Set11 BSD68 Urban100 全模块 区域分支 全局分支 × √ — — 40.34 36.66 37.31 √ — √ × 40.44 36.49 37.24 √ — × × 40.28 36.30 37.15 √ × — — 39.80 35.56 36.60 √ √ √ √ 40.62 36.93 37.48
Tools
Get Citation
Copy Citation Text
Yuhong LIU, Heng YANG. Iterative reconstruction of compressive sensing combining image hierarchical-feature[J]. Optics and Precision Engineering, 2024, 32(14): 2311
Paper Information
Category:
Received: Mar. 1, 2024
Accepted: --
Published Online: Sep. 27, 2024
The Author Email: YANG Heng (11220697@stu.lzjtu.edu.cn)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20