Denoising by hard-thresholding in 3D transform domain with block-matching

  • Method. For all blocks $Z_{x_{R}\in X}$ from the noisy image, we do the following steps.

  • Block-matching. The result of block-matching is a set MATH of the coordinates of the blocks that are similar to the currently-processed reference block, $Z_{x_{R}}$, according to our $d$-distance measure;

  • Denoising by hard-thresholding in local 3D transform domain.MATH

    • MATH comprises of MATH stacked local estimates MATH of the true image blocks located at the matched locations MATH.

    • $\omega _{x_{R}}$ is a weight which is inversely proportional to the number of non-zero transform coefficients after hard-thresholding, $N_{har}$.