Wiener filtering in 3D transform domain with block-matching

  • Wiener filtering approach. We develop an empirical Wiener filtering extension provided an estimate of the noise-free image is available, obtained from the hard-thresholding technique, for example.

  • Block-matching. We do block-matching among blocks extracted from the initial estimate. Thus, we replace the $d$-distance measure with the normalized $L^{2}$-norm of the difference of two blocks with subtracted means; the definition of MATH becomes

  • Wiener filtering in local 3D transform domain.

    • Create a 3D array of Wiener filter's attenuating coefficients

    • Produce local estimates and their weight:MATHin analogy to the hard-thresholding approach.