Optimization BM3D Algorithm Based on Pseudo-3D Transform

A optimization BM3D algorithm based on pseudo-3D transformation (incomplete 3D transformation) is proposed to solve the problem of huge computational of BM3D image denoising algorithm. Firstly, one-dimensional wavelet transform was carried out on the third dimension of 3D similar blocks to get one-dimensional coefficient blocks. Then the lowest-frequency coefficient block gotten above was transformed by two-dimensional transformation. Following, two-dimensional inverse transformation was performed on the lowest-frequency coefficient blocks after hard threshold filtering or Wiener filtering. And finally, one-dimensional inverse transformation was carried out on the third dimension to get the noise reduced image. The performance of the proposed optimization method is very closed to the original BM3D algorithm, but the computation of pseudo-3D transformation is only 1/3 of that of 3D transformation.