Fig. 1. Process of two-dimensional wavelet transform for image I(x, y)

Fig. 2. Effects of two-dimensional DWT for butterfly_GT.bmp in Set5

Fig. 3. Flowchart of the proposed method

Fig. 4. Network architecture of DRRN. (a) Hole architecture; (b) structure of recursive block

Fig. 5. Structure of modified recursive block. (a) Removing BN layers; (b) decreasing convolution layers

Fig. 6. Realization schematic of the proposed method

Fig. 7. Whole and local comparisons of “monarch.bmp” in Set14 processed with scale ×3. (a) Real image; (b) Bicubic/29.43 dB; (c) SRCNN/32.39 dB; (d) FSRCNN/32.44 dB; (e) VDSR/34.51 dB; (f) DRRN/34.68 dB; (g) proposed SDN/34.53 dB; (h) proposed CWTDN/34.82 dB

Fig. 8. Whole and local comparisons of “21077.bmp” in BSD100 processed with scale ×4. (a) Real image; (b) Bicubic/24.88 dB; (c) SRCNN/25.84 dB; (d) FSRCNN/26.13 dB; (e) VDSR/26.57 dB; (f) DRRN/26.75 dB; (g) proposed SDN/26.60 dB; (h) proposed CWTDN/26.91 dB