mrpro.algorithms.denoiser.transform_domain_soft_thresholding

mrpro.algorithms.denoiser.transform_domain_soft_thresholding(idata: IData, transform_op: LinearOperator, threshold: float | Tensor) → IData
mrpro.algorithms.denoiser.transform_domain_soft_thresholding(idata: Tensor, transform_op: LinearOperator, threshold: float | Tensor) → Tensor

Apply domain thresholding.

This algorithm implements denoising by applying :math:x = W^H mathrm{SoftThreshold}_l (W y), where :math:y is the given noisy image, :math:W is a linear operator, and :math:mathrm{SoftThreshold}_l is the soft-thresholding function with parameter :math:l>0.

This operator is well-known in the signal-processing community under the name of “Domain Transform Thresholding” and has various applications and interpretations, see for example [AHMAD2020]. For instance, when :math:W is orthogonal, the algorithm gives the closed-form solution of the problem :math:min_x frac{1}{2}||x - y||_2^2 + l ||Wx||_1.

References

[AHMAD2020]

Ahmad R, Bouman CA, Buzzard GT, Chan S, Liu S, Reehorst ET, Schniter P (2020) Plug-and-Play Methods for Magnetic Resonance Imaging: Using Denoisers for Image Recovery. IEEE Signal Process Mag 37(1):105-116. https://doi.org/10.1109/msp.2019.2949470

Parameters:
  • idata (IData | Tensor) – input image

  • transform_op (LinearOperator) – linear operator representing the transformation from image space to transform space, in which the coefficients will be thresholded

  • threshold (float | Tensor) – Threshold for soft-thresholding

Returns:

the thresholded image.