Improved DCT-Based Nonlocal Means Filter for MR Images Denoising
Figure 4
The comparison of the weights distribution used to estimate the central pixel of the left images with the noise-free and noisy conditions between the NLM algorithm and the proposed algorithm. (b) and (e) are the similarity weights distribution images for (a) and (d), respectively, which are calculated by the NLM filter in neighborhoods’ full space of the central pixel. (c) and (f) are also the similarity weights distribution images for (a) and (d), correspondingly, which are obtained by the proposed filter in neighborhoods’ DCT subspace of the central pixel. The weights distribution images are displayed with the logarithmic scale.