Author Affiliations
1School of Electronics and Information, Xi'an Polytechnic University, Xi'an , Shaanxi 710048, China2School of Vocational and Technical Education, Guangxi Science & Technology Normal University, Laibin , Guangxi 546199, Chinashow less
Fig. 1. Framework of the proposed method
Fig. 2. Block diagram of Boosting regressor learning algorithm
Fig. 3. Block diagram of refinement stage
Fig. 4. Performance comparison of different super-resolution methods on Set5 dataset for ×3 magnification
Fig. 5. Super-resolution results of “Flower” in Set10 dataset for ×3 magnification. (a) Original image; (b) A+ method; (c) SRCNN method; (d) Zhang's method; (e) MMPM method; (f) Ours
Fig. 6. Super-resolution results of “Img092” in Urban100 dataset for ×3 magnification. (a) Original image; (b) A+ method; (c) SRCNN method; (d) Zhang's method; (e) MMPM method; (f) Ours
Fig. 7. Super-resolution results on real-world dataset for ×3 magnification. (a) A+ method; (b) SRCNN method; (c) Zhang's method; (d) MMPM method; (e) Ours
Fig. 8. Influence of parameter T on the average PSNR in B100 dataset
Fig. 9. Influence of parameter T on the average PSNR in Set10 dataset
Fig. 10. Influence of sub-dictionary size on the reconstruction results in Set5 dataset. (a) Average PSNR value varing with sub-dictionary size; (b) average SSIM value varing with sub-dictionary size
Dataset | Evaluation index | A+ | SRCNN | Zhang’s | MMPM | Ours (K=512) | Ours (K=1024) |
---|
Set5 | PSRN | 36.41 | 36.35 | 36.19 | 36.80 | 36.77 | 36.81 | SSIM | 0.951 | 0.952 | 0.950 | 0.956 | 0.955 | 0.955 | Set10 | PSRN | 33.00 | 32.94 | 32.76 | 33.43 | 33.50 | 33.56 | SSIM | 0.927 | 0.927 | 0.925 | 0.933 | 0.933 | 0.934 | Set14 | PSRN | 32.22 | 32.23 | 32.07 | 32.43 | 32.50 | 32.53 | SSIM | 0.902 | 0.904 | 0.899 | 0.907 | 0.907 | 0.907 | B100 | PSRN | 31.09 | 31.13 | 30.94 | 31.35 | 31.35 | 31.38 | SSIM | 0.881 | 0.884 | 0.877 | 0.889 | 0.889 | 0.889 | Urban100 | PSRN | 28.98 | 29.07 | 28.97 | 29.51 | 29.64 | 29.69 | SSIM | 0.886 | 0.889 | 0.886 | 0.898 | 0.900 | 0.900 |
|
Table 1. Average PSNR and SSIM values of different methods for ×2 magnification
Dataset | Evaluation index | A+ | SRCNN | Zhang’s | MMPM | Ours (K=512) | Ours (K=1024) |
---|
Set5 | PSRN | 32.46 | 32.41 | 32.56 | 32.66 | 32.76 | 32.78 | SSIM | 0.905 | 0.904 | 0.907 | 0.910 | 0.911 | 0.911 | Set10 | PSRN | 29.04 | 29.07 | 29.11 | 29.25 | 29.40 | 29.48 | SSIM | 0.845 | 0.843 | 0.848 | 0.852 | 0.855 | 0.856 | Set14 | PSRN | 29.09 | 29.02 | 29.10 | 29.19 | 29.22 | 29.25 | SSIM | 0.816 | 0.814 | 0.816 | 0.821 | 0.819 | 0.820 | B100 | PSRN | 28.16 | 28.19 | 28.22 | 28.33 | 28.35 | 28.38 | SSIM | 0.775 | 0.779 | 0.779 | 0.786 | 0.785 | 0.785 | Urban100 | PSRN | 25.93 | 25.85 | 26.02 | 26.10 | 26.25 | 26.28 | SSIM | 0.792 | 0.787 | 0.794 | 0.798 | 0.804 | 0.806 |
|
Table 2. Average PSNR and SSIM values of different methods for ×3 magnification
R | Set5 | Set10 | Set14 | B100 | Urban100 |
---|
PSNR | SSIM | PSNR | SSIM | PSNR | SSIM | PSNR | SSIM | PSNR | SSIM |
---|
1 | 32.610 | 0.9081 | 29.197 | 0.8508 | 29.133 | 0.8178 | 28.284 | 0.7826 | 25.998 | 0.7956 | 2 | 32.745 | 0.9103 | 29.368 | 0.8544 | 29.217 | 0.8193 | 28.349 | 0.7843 | 26.199 | 0.8018 | 3 | 32.759 | 0.9107 | 29.403 | 0.8552 | 29.219 | 0.8194 | 28.350 | 0.7845 | 26.258 | 0.8039 |
|
Table 3. Performance evaluation on five benchmarks by using different cascaded times