Table 10 Quality analysis of recovered image after denoising using CNN.
From: Convolutional neural network and wavelet composite against geometric attacks a watermarking approach
Noise | SSIM | NC |
|---|---|---|
Gaussian filtering (3 × 3) | 0.9991 | 0.998 |
Gaussian filtering (5 × 5) | 0.9999 | 0.999 |
Gaussian filtering (9 × 9) | 0.9992 | 0.997 |
Median filtering (3 × 3) | 0.9999 | 0.998 |
Median filtering (5 × 5) | 0.9996 | 0.997 |
Average filtering (3 × 3) | 0.9997 | 0.998 |
Average filtering (5 × 5) | 0.9996 | 0.998 |
JPEG compression (Q = 30) | 0.9998 | 0.998 |
JPEG compression (Q = 60) | 0.9993 | 0.998 |
JPEG compression (Q = 90) | 0.9949 | 0.945 |
Contrast adjustment | 0.9997 | 0.998 |
Histogram equalisation | 0.9996 | 0.998 |
Rotation attacks 30(degree) | 0.9994 | 0.998 |
Rotation attacks 60(degree) | 0.9995 | 0.998 |
Rotation attacks 90(degree) | 0.9 946 | 0.922 |
Image cropping 25% | 0.9996 | 0.998 |
Image cropping 35% | 0.9923 | 0.948 |
Image scaling (Shrinking) | 0.9993 | 0.998 |
Image scaling 1.2 | 0.9997 | 0.998 |
Image scaling 1.5 | 0.9992 | 0.998 |
Gaussian noise (m = 0,v = 0.01) | 0.9999 | 0.998 |
Gaussian noise (m = 0,v = 0.05) | 0.9996 | 0.998 |
Salt and pepper noise (d = 0.01) | 0.9997 | 0.998 |
Salt and pepper noise (d = 0.05) | 0.9994 | 0.998 |
Speckle noise (v = 0.2) | 0.9992 | 0.998 |
Speckle noise (v = 0.6) | 0.9991 | 0.998 |
