Fig 1.
The flowchart of the proposed SR algorithm.
Fig 2.
γ versus average PSNR on Set5.
(A) upscaling factor ×2; (B) upscaling factor ×3; (C) upscaling factor ×4.
Fig 3.
Dictionary size influence on performance on average on Set5.
(A) upscaling factor ×2; (B) upscaling factor ×3; (C) upscaling factor ×4.
Fig 4.
Number of similar atoms influence on performance on average on Set5.
(A) upscaling factor ×2; (B) upscaling factor ×3; (C) upscaling factor ×4.
Table 1.
Comparisons of average PSNR (dB) and SSIM (σ = 0).
Table 2.
The results of PSNR (dB) and SSIM on the set5 dataset.
Table 3.
The results of average PSNR (dB) and SSIM on the Set14 and B100 dataset.
Fig 5.
Comparisons with various image super-resolution methods on “coastguard” from Set14 with upscaling factor ×2 (σ = 0, PSNR in dB).
(A) Ground truth HR; (B) NE [22]; (C) SCSR [25]; (D) Zedye [26]; (E) A+ [31]; (F) SRCNN [20]; (G) CSC [32]; (H) ours.
Fig 6.
Comparisons with various image super-resolution methods on “16077” from B100 with upscaling factor ×2 (σ = 10, PSNR in dB).
(A) Ground truth HR; (B) NE [22]; (C) SCSR [25]; (D) Zedye [26]; (E) A+ [31]; (F) SRCNN [20]; (G) CSC [32]; (H) ours.
Fig 7.
Comparisons with various image super-resolution methods on “241004” from B100 with upscaling factor ×3 (σ = 10, PSNR in dB).
(A) Ground truth HR; (B) NE [22]; (C) SCSR [25]; (D) Zedye [26]; (E) A+ [31]; (F) SRCNN [20]; (G) CSC [32]; (H) ours.
Fig 8.
Comparisons with various image super-resolution methods on “208001” from B100 with upscaling factor ×4 (σ = 10, PSNR in dB).
(A) Ground truth HR; (B) NE [22]; (C) Zedye [26]; (D) A+ [31]; (E) SRCNN [20]; (F) CSC [32]; (G) ours.
Table 4.
Comparisons of average reconstruction time (s)on Set5.
Table 5.
Effect of IBP on average PSNR(dB) and SSIM (Set 5).
Fig 9.
Effect of distance penalty on average PSNR (dB)(Set 5).
(A) upscaling factor ×2; (B) upscaling factor ×3; (C) upscaling factor ×4.