Research Article
Pixelwise Estimation of Signal-Dependent Image Noise Using Deep Residual Learning
Table 1
Average error on simulated homogeneous noise (dB).
| Dataset | Noise level | Pyatykh [24] | Liu [25] | Chen [7] | DRNE |
| Kodak (24 images) | | 0.91 | 0.96 | 0.68 | 0.74 | | 1.79 | 0.17 | 0.09 | 0.08 | | 3.93 | 0.19 | 0.12 | 0.18 | | 6.13 | 0.33 | 0.29 | 0.29 | | 8.42 | 0.55 | 0.52 | 0.51 | | 10.87 | 0.90 | 0.85 | 0.80 |
| McMaster (18 images) | | 0.62 | 0.49 | 1.60 | 0.83 | | 1.70 | 0.16 | 0.23 | 0.11 | | 4.12 | 0.65 | 0.32 | 0.41 | | 6.93 | 1.22 | 0.81 | 0.91 | | 9.84 | 1.89 | 1.51 | 1.53 | | 12.73 | 2.65 | 2.17 | 2.26 |
| BSD500 (500 images) | | 0.52 | 0.35 | 0.54 | 0.32 | | 1.66 | 0.18 | 0.05 | 0.13 | | 3.70 | 0.35 | 0.21 | 0.32 | | 5.93 | 0.57 | 0.46 | 0.53 | | 8.30 | 0.90 | 0.79 | 0.85 | | 10.79 | 1.33 | 1.23 | 1.26 |
|
|
Bold fonts denote the best performance and italics denotes the second best performance.
|