pySPFM._solvers.spatial_regularization.spatial_structured_sparsity#
- spatial_structured_sparsity(estimates, data, mask, niter, dims, lambda_)[source]#
Spatial regularization technique based on the structured sparsity as in Total Activation.
This function computes the structured sparsity regularization and is another variant of fgp algorithm for structured sparsity
solves \(\\frac{1}{2} \\| \\mathbf{y} - \\mathbf{x} \\|^2 + \\lambda \\| \\mathbf{D}^\\textrm{order} \\cdot \\mathbf{x} \\|_{s,2,1}\).
Delta is the laplacian operator delta[n] = [1 -2 1]; so symmetric, in matrix form \(\\Delta^T = \\Delta\).
- Parameters:
estimates (ndarray) – Estimates (output of temporal regularization).
data (ndarray) – Observations.
mask (Nibabel object) – Mask image to unmask and mask estimates (2D to 4D and back).
niter (int) – Number of iterations to perform spatial regularization.
dims (list) – Dimensions of the data.
lambda_ (float) – Spatial regularization parameter.
- Returns:
final_estimates (ndarray) – Estimates of activity-inducing or innovation signal after spatial regularization.