From: Improved BM3D image denoising using SSIM-optimized Wiener filter
 |  | Fast profile | Normal profile | ||
---|---|---|---|---|---|
 | Notations | Meaning |  | σ≤40 | σ>40 |
Parameters for | \(\tau ^{ht}_{2D}\) | 2D transform used | 2D-Bior1.5 | 2D-Bior1.5 | 2D-DCT |
step 1 (ht) | \(N^{ht}_{1}\) | Patch size | 8 | 8 | 12 |
 | \(N^{ht}_{2}\) | Maximum number of | 16 | 16 | 16 |
 |  | similar patches retained |  | ||
 | \(N^{ht}_{step}\) | Step of ref. patch | 6 | 3 | 4 |
 | \(N^{ht}_{S}\) | Size of search window | 25 | 39 | 39 |
 | \(N^{ht}_{FS}\) | Exhaustive search window size | 6 | 1 | 1 |
 | \(N^{ht}_{PR}\) | Predictive search window size | 3 | - | - |
 | β ht | Parameters for Kaiser window | 2.0 | 2.0 | 2.0 |
 | λ 2D | Pre-processing threshold | 0 | 0 | 2.0 |
 | λ 3D | Hard threshold | 2.7 | 2.7 | 2.8 |
 | \(\tau ^{ht}_{match}\) | Similarity threshold for patches | 2500 | 2500 | 5000 |
Parameters for | \(\tau ^{wie}_{2D}\) | 2D transform used | 2D-DCT | 2D-DCT | 2D-DCT |
step 2 (wie) | \(N^{wie}_{1}\) | Patch size | 8 | 8 | 11 |
 | \(N^{wie}_{2}\) | Maximum number of | 16 | 32 | 32 |
 |  | similar patches retained |  | ||
 | \(N^{wie}_{step}\) | Step of ref. patch | 5 | 3 | 6 |
 | \(N^{wie}_{S}\) | Size of search window | 25 | 39 | 39 |
 | \(N^{wie}_{FS}\) | Exhaustive search window size | 5 | 1 | 1 |
 | \(N^{wie}_{PR}\) | Predictive search window size | 2 | - | - |
 | \(\tau ^{wie}_{match}\) | Similarity threshold for patches | 400 | 400 | 3500 |
 | β wie | Parameters for Kaiser window | 2.0 | 2.0 | 2.0 |
Common | Â | Â | 1D-Haar | 1D-Haar | 1D-Haar |